add sessions/0068/SESSION-0068.0-TRANSCRIPT-2026-07-03T06-53--2026-07-03T07-39.md + replace placeholder/variant SESSION-0068.0-TRANSCRIPT-2026-07-03T06-53--INPROGRESS.md

add sessions/0067/SESSION-0067.0-TRANSCRIPT-2026-07-02T20-07--2026-07-03T07-39.md + replace placeholder/variant SESSION-0067.0-TRANSCRIPT-2026-07-02T20-07--INPROGRESS.md
claim vscode-cowriting-plugin session 0068 (placeholder) + sessions.json entry
2026-07-03 07:42:11 -07:00 · 2026-07-03 07:42:07 -07:00 · 2026-07-03 06:54:04 -07:00 · 2026-07-02 20:21:06 -07:00 · 2026-07-03 03:20:23 +00:00 · 2026-07-02 20:08:20 -07:00
152 changed files with 26664 additions and 717 deletions
@@ -8,3 +8,6 @@ test/e2e/fixtures/workspace/.threads/

 # Sandbox playground churn (the EDH workspace - play freely, commit nothing)
 sandbox/.threads/
+
+# brainstorming visual-companion scratch
+.superpowers/
@@ -14,7 +14,59 @@ catalog (a pure, key-free SDK call) in a notification and the
 Features shipped so far: F2 region-anchored threads (Feature #4), F3 live
 human/Claude attribution (Feature #6), F4 propose/accept diff flow
 (Feature #12), F5 cross-rung sidecar contract (Feature #14), F6 diff-view
-toggle (Feature #17).
+data layer (Feature #17), and the **native-surfaces migration** (the
+"Coediting Markdown with a Machine" Solution Design) — the current UI, replacing
+the bespoke F7/F10/F11 review webview with VS Code's own chrome end to end. See
+**How it works** below for the current experience; the F7/F9/F10/F11 sections
+further down are kept as a **historical record** of the superseded webview (each
+now says so up top).
+
+## How it works
+
+There is **no bespoke panel**. Every surface below is native VS Code chrome —
+the editor title bar, status bar, Comments gutter, source control, diff editor,
+and the built-in Markdown preview:
+
+1. **Enter coediting.** Run **"✦ Coedit this Document with Claude"** (command
+   palette, editor title menu, or editor/explorer context menu) on a Markdown
+   document — it captures a **baseline** (a clean checkpoint: git HEAD if the
+   file is tracked and clean, otherwise a snapshot) to measure future changes
+   against. Until a document is entered, none of the surfaces below attach to
+   it. **"Stop editing with Claude"** reverses it.
+2. **Title bar + status bar are the everyday entry points.** Once coedited, the
+   editor title bar carries **Review Changes**, **Ask Claude**, **Toggle
+   Annotations**, and (in snapshot mode) **Mark Changes as Reviewed**; a
+   `✦ Coediting · N changes` status-bar item shows the pending-change count and
+   opens the review when clicked — no need to find the Source Control pane.
+3. **Ask Claude is comments-first.** **"Ask Claude"** (editor title, or select
+   text → right-click → "Ask Claude to Edit") opens a focused comment box on
+   the native **Comments API** — on your selection, or a top-anchored
+   whole-document thread with none selected. Claude replies in-thread and
+   offers an edit; accepting the offer turns it into one or more pending
+   proposals.
+4. **Pending changes live in the buffer.** A proposal is **optimistically
+   applied** straight into the editor (green/blue insertion tint, a struck
+   deletion hint) with an inline **`✓ Keep` / `✗ Reject`** CodeLens above each
+   changed block — and, once ≥2 proposals are pending, a top-of-file
+   **`✓ Keep all (N)` / `✗ Reject all`** pair. Nothing is force-applied: a
+   proposal whose target text changed underneath it is skipped, never guessed.
+5. **Native diff answers "what changed?"** **Review Changes** (editor title, or
+   `Ctrl+Alt+R` / `Cmd+Alt+R`) opens VS Code's own diff editor — baseline on the
+   left, your live document on the right — with the platform's inline/side-by-side
+   toggle, navigation, and accessibility for free. A Source Control gutter/viewlet
+   entry (quick-diff change bars, "Open Changes") is a bonus home for git users,
+   never the required path.
+6. **The built-in Markdown preview is the annotated read.** "Open Preview to the
+   Side" (or right-click a markdown file/tab → **"Open Cowriting Review
+   Preview"**, `cowriting.openReviewPreview`) renders the document with
+   authorship/change coloring **inside VS Code's own preview** — green = human,
+   blue = Claude, strikethrough = deleted — toggled by **Toggle Annotations**
+   (command or the title-bar eye icon). No separate webview, no separate
+   persistence.
+
+Design: `vscode-cowriting-plugin-content/specs/coauthoring-native-surfaces.md`
+(§5 UX Layout is the canonical source for this map). Migration plan:
+[`docs/superpowers/plans/2026-07-01-native-surfaces-migration.md`](docs/superpowers/plans/2026-07-01-native-surfaces-migration.md).

 ## Architecture

@@ -53,6 +105,14 @@ Design: `vscode-cowriting-plugin-content/specs/coauthoring-inner-loop.md`. No li

 ## F3 — Live human/Claude attribution (Feature #6)

+> **Commands superseded (native-surfaces migration).** The attribution **data
+> layer** below is current, but the Commands block is historical:
+> `Cowriting: Toggle Attribution` was retired — authorship/change coloring now
+> toggles via **Toggle Annotations** on a coedited document (How it works §6) —
+> and "Ask Claude to Edit Selection" is palette-hidden; use **Ask Claude to
+> Edit** (`Ctrl+Alt+E` / `Cmd+Alt+E`) or the comments-first **Ask Claude** (How
+> it works §3). Kept below as a historical record.
+
 As you and Claude coauthor, every span in the buffer carries an author: human
 edits render with a subtle left border, Claude-authored spans with a background
 tint. Text that predates tracking stays plain — the honest record. Edits that
@@ -132,7 +192,16 @@ record per the contract; git push/pull is the transport, no re-homing ever.

 Design: `vscode-cowriting-plugin-content/specs/coauthoring-cross-rung-format.md`.

-## F6 — Diff-view toggle (Feature #17, #19)
+## F6 — Diff-view toggle (Feature #17, #19) — SUPERSEDED
+
+> **Superseded.** The two-pane `Ctrl+Alt+D` toggle this section describes was
+> deleted in #34; only the baseline **data layer** survived, now serving the
+> native diff editor (`Ctrl+Alt+R` / "Review Changes" — How it works §5) and the
+> built-in preview annotations (§6). The native-surfaces migration's Task 2
+> additionally **retired INV-18** (machine-landing auto-advance): a landed
+> Claude edit stays a visible change until "Mark Changes as Reviewed" (the
+> renamed "Pin Diff Baseline to Now"), it no longer auto-clears. Kept below as a
+> historical record.

 **`Ctrl+Alt+D`** (the same chord on macOS — not `Cmd`; or **Cowriting: Toggle
 Diff View**) flips the focused document into a native `vscode.diff` against a
@@ -164,6 +233,158 @@ instead of git archaeology.
 Design: `vscode-cowriting-plugin-content/specs/coauthoring-diff-view.md`.
 Live smoke: [`docs/MANUAL-SMOKE-F6.md`](docs/MANUAL-SMOKE-F6.md).

+## F7 — Rendered track-changes preview (Feature #21) — SUPERSEDED
+
+> **Superseded (Task 8, native-surfaces migration).** The bespoke webview this
+> section describes was deleted; its authorship/change coloring lives on
+> **inside VS Code's own built-in Markdown preview** — see **How it works** §6
+> above. Kept below as a historical record of the pre-migration design. (The
+> `Ctrl+Alt+R` chord below now opens the native **Review Changes** diff; the
+> annotated preview opens via **Open Cowriting Review Preview**, no chord.)
+
+**`Ctrl+Alt+R`** (or **Cowriting: Open Track-Changes Preview**) opens a
+read-only webview **beside** a **Markdown** editor that renders the document and
+marks what changed since the F6 baseline — the "track changes" / "suggesting
+mode" altitude rather than a raw-text split-diff:
+
+- **Prose** additions are highlighted (`<ins>`), deletions struck (`<del>`),
+  refined to the word.
+- **Code and mermaid fences** are diffed **whole** (atomic, INV-23): a changed or
+  added one renders fully with a small **"changed"** badge; a removed one renders
+  struck. Mermaid fences render as **diagrams** (mermaid runs in the webview).
+  Intra-diagram node/edge diffing is deferred (#22).
+- It **updates live** as you and Claude edit (debounced), and **re-bases** when
+  Claude lands an edit (baseline advances, INV-18) or you pin — so accepted text
+  drops its marks. It **reuses the F6 baseline** and adds **no** persistence
+  (pure read-only, INV-20).
+- The webview is **sealed** (INV-21): local bundled assets only, strict CSP with
+  a per-load nonce, no network/CDN, no LLM. Mermaid is bundled into the
+  **webview** asset only, never the extension-host bundle.
+- The render engine (`src/trackChangesModel.ts`) is a **pure, vscode-free**
+  function (INV-22), unit-tested with no editor and no webview; host E2E
+  (`test/e2e/suite/trackChangesPreview.test.ts`) drives the same programmatic
+  propose/accept seam with no LLM.
+
+F7 is **markdown-only**; for any other file (incl. code), use F6's diff toggle
+(`Ctrl+Alt+D`). The webview's visual rendering (mermaid, theming) is verified by
+the manual smoke, not the sealed-sandbox E2E.
+
+Design: `vscode-cowriting-plugin-content/specs/coauthoring-rendered-preview.md`.
+Live smoke: [`docs/MANUAL-SMOKE-F7.md`](docs/MANUAL-SMOKE-F7.md).
+
+## F9 — Authorship view in the preview (Feature ~#27) — SUPERSEDED
+
+> **Superseded (Task 8, native-surfaces migration).** The F7 webview this mode
+> lived in was deleted; the built-in preview's annotations (How it works §6)
+> always show authorship + change coloring together — there is no separate
+> mode toggle. Kept below as a historical record.
+
+The rendered preview (F7) gains a second mode, switched by a `[ Track changes |
+Authorship ]` toggle in its header. **Authorship** mode re-renders the current
+document with each span colored by its F3 author — Claude (blue) vs you (green),
+inline and char-precise — with a legend. Unlike track-changes (which diffs against
+the F6 baseline, and so hides Claude's text once the baseline advances past a
+landing), authorship reads F3 attribution directly, so Claude's contributions stay
+visible. Code/mermaid fences carry a block-level author badge (atomic). Read-only,
+sealed webview, no new persistence (INV-26..28).
+
+Design: `docs/superpowers/specs/2026-06-11-authorship-preview-design.md`.
+Live smoke: [`docs/MANUAL-SMOKE-F9.md`](docs/MANUAL-SMOKE-F9.md).
+
+## F8 — Out-of-workspace authoring (Feature #25)
+
+"Ask Claude to Edit Selection" (and F2 threads / F3 attribution / F4
+propose-accept) now work on **any** document the editor shows — saved in the
+workspace folder, saved outside it, or untitled — matching the already-universal
+F6 diff and F7 preview. Authoring is no longer gated to in-workspace files, and
+the commands are live even with **no folder open**.
+
+Persistence is **hybrid** (one `SidecarStore` abstraction, routed per-document):
+
+- an **in-workspace** file keeps its committable `.threads/<repo-rel>.json`
+  sidecar, **byte-for-byte unchanged** (INV-2) — the only home the F5 cross-rung
+  contract ever sees;
+- an **out-of-workspace** file or **untitled** buffer stores its coauthoring
+  artifact in VS Code **global storage** keyed by `sha256(uri)` (the same home
+  and key F6's baseline uses, INV-19/24); untitled buffers are **in-memory
+  only** (lost on reload/save).
+
+A global-storage artifact is **not a committed file**, so it is **never
+cross-rung-shareable** (INV-25), and renaming/moving the file **orphans** its
+artifact (the key is the URI hash) — both stated as design contract, not
+discovered later. Routing leaves the in-workspace path untouched, so rollback is
+a plain PR revert with zero data migration.
+
+Design: `vscode-cowriting-plugin-content/specs/coauthoring-out-of-workspace.md`.
+Live smoke: [`docs/MANUAL-SMOKE-F8.md`](docs/MANUAL-SMOKE-F8.md).
+
+## F10 — Interactive review: write left / review right (Feature #29) — SUPERSEDED
+
+> **Superseded (Task 8, native-surfaces migration).** The editor is no longer
+> "zero-annotation" — pending proposals now render optimistically applied
+> in-buffer with `✓ Keep`/`✗ Reject` CodeLens (F12), and review happens across
+> the native diff editor + built-in preview, not one dedicated webview. See
+> **How it works** above. Kept below as a historical record. (The `Ctrl+Alt+R`
+> chord below now opens the native **Review Changes** diff, not this panel.)
+
+A clean, **zero-annotation editor** on the left; the rendered preview on the
+right as the **single interactive review surface**. The editor carries no
+attribution tint, no in-editor proposal threads, and no diff — all review lives
+in the preview, toggled by the **Annotations** switch in its header (on by
+default). Open it via `Ctrl+Alt+R`, the editor title-bar button, or
+**right-click a markdown file in the Explorer / its editor tab →
+"Open Cowriting Review Panel"** (#41) — the right-click acts on the clicked
+document, opening it first if needed.
+
+In the on-state the preview shows **green = human / blue = Claude /
+strikethrough = deleted**, and surfaces each of Claude's pending F4 proposals as
+a blue `cw-proposal` block with **✓ / ✗** buttons: **✓** accepts (the
+replacement lands Claude-attributed via the seam and the baseline advances past
+it), **✗** rejects (the block vanishes, the document untouched). With no preview
+open, a status-bar indicator shows the pending-proposal count and opens the
+review when clicked. Toggle **Annotations** off for clean rendered markdown.
+Read-only, sealed webview, no new persistence (INV-32..34).
+
+F6's two-pane diff and F9's authorship view are **retained only as data layers**
+(the baseline the preview diffs against; the F3 attribution that colors it) —
+they are no longer separate user surfaces.
+
+Design: `vscode-cowriting-plugin-content/specs/coauthoring-interactive-review.md`.
+Live smoke: [`docs/MANUAL-SMOKE-F10.md`](docs/MANUAL-SMOKE-F10.md).
+
+## F11 — Preview toolbar as the primary interaction surface (Feature #43) — SUPERSEDED
+
+> **Superseded (Task 8, native-surfaces migration).** The webview toolbar this
+> section describes was deleted; its two controls moved to native chrome —
+> **Pin baseline** is now **"Mark Changes as Reviewed"** in the editor title
+> bar, and **Ask Claude** is the comments-first ask (How it works §3). Kept
+> below as a historical record.
+
+The review preview's **header toolbar** becomes the cockpit for the inner loop.
+Beside the existing **Annotations** switch it gains two controls:
+
+- **⌖ Pin baseline** — pins the previewed document's review baseline to now and
+  clears the change-marks (homes the previously-orphaned `pinDiffBaseline`
+  command, which is reachable from the palette again too).
+- **✦ Ask Claude…** — one **adaptive** button. Its label flips on the preview's
+  selection: **Edit Selection** when text is selected in the rendered preview,
+  **Edit Document** otherwise. Clicking it opens a host input box for the
+  instruction (the LLM turn and prompt stay host-side — the sealed webview gains
+  no LLM/credential surface), then surfaces the result as F4 proposals — **one**
+  for a selection (mapped back to its source block-union), or **one per changed
+  hunk** for a whole-document rewrite (`diffToHunks`), each independently ✓/✗-able.
+
+The pure render layer emits `data-src-start`/`data-src-end` on every block in
+**both** modes (INV-36); the webview's only mapping duty is walking a selection
+to its nearest `data-src` ancestor. Right-clicking a markdown tab → **Open Review
+Preview** is the minimal gateway making the surface reachable end to end (#41/#42
+expand it). Edit controls are disabled on a non-authorable doc (reading stays
+allowed). No new model, no new persistence — pin via the F6 store, edits via the
+F4 propose/accept seam with F3 attribution (INV-35..37).
+
+Design: [`docs/superpowers/specs/2026-06-12-f11-preview-toolbar-interaction-surface.md`](docs/superpowers/specs/2026-06-12-f11-preview-toolbar-interaction-surface.md).
+Live smoke: [`docs/MANUAL-SMOKE-F11.md`](docs/MANUAL-SMOKE-F11.md).
+
 ## Develop

 - `npm run watch` — rebuild on change.
@@ -0,0 +1,58 @@
+# Manual smoke — F10 interactive review in the preview (#29)
+
+F10 makes the rendered preview the **single interactive review surface**: the
+editor is clean (zero annotations), and you accept/reject Claude's proposals
+*inside the preview*. The webview's *visual* rendering (theming, ✓/✗ buttons) is
+verified here, not in the automated host E2E (the webview is a sealed sandbox).
+Run once per change that touches F10. One live turn hits the SDK (or use the
+`proposeAgentEdit` seam to stay key-free).
+
+## Setup
+
+1. `npm run build`
+2. Launch the Extension Development Host (F5 in VS Code, or the Run panel) with
+   `sandbox/` open.
+3. Open a markdown document containing some prose (e.g. copy
+   `test/e2e/fixtures/workspace/docs/preview.md`).
+
+## Steps
+
+1. **Clean editor.** Look at the source editor: there is **no attribution tint,
+   no in-editor proposal comment threads, and no diff** — the editor is a plain
+   text buffer (F10/INV-32). All review lives in the preview.
+2. **Open the preview.** Run **"Cowriting: Open Review Preview"** (or
+   `Ctrl+Alt+R`). A preview opens beside the editor. The header shows an
+   **Annotations** switch (on by default) and a summary.
+3. **Edit prose.** In the source editor, change a word in a paragraph. The
+   preview updates (≈150 ms) in its on-state: the new word highlighted as a green
+   insertion (`<ins>` / `cw-by-human`), the old word struck (`<del>` / `cw-del`);
+   the summary increments. Your own typing is colored green (human).
+4. **Ask Claude to edit a selection.** Select a sentence → **"Ask Claude to Edit
+   Selection"** → instruct (or invoke the `proposeAgentEdit` seam). A **blue
+   `cw-proposal` block** appears in the preview, showing the struck replaced text
+   and the proposed replacement, with **✓ / ✗** buttons in a `cw-actions` span.
+   The editor itself does **not** change (INV-10 — propose never mutates the doc).
+5. **Accept one.** Click **✓** on the proposal. Expect: the replacement **lands
+   in the document** (the editor text updates), the proposal block **clears** from
+   the preview, and the landed Claude text is **not** marked as a change (the
+   baseline advanced past the landing; INV-18).
+6. **Reject another.** Propose a second edit, then click **✗** on it. Expect: the
+   block **vanishes** from the preview and the **document is unchanged**.
+7. **Toggle Annotations off.** Flip the header **Annotations** switch off. Expect:
+   the preview shows **clean rendered markdown** — no green/blue author colors, no
+   struck deletions, no proposal blocks (INV-33). Flip it **on** again: the marks
+   and any pending proposal blocks return.
+8. **Status-bar indicator (PUC-6).** Close the preview. With a **pending
+   proposal** outstanding and **no preview open**, a status-bar item shows the
+   pending count (e.g. "1 Claude proposal"). **Click it** — the review preview
+   opens and the indicator disappears.
+9. **Theme.** Toggle light / dark / high-contrast (`Ctrl+K Ctrl+T`). The proposal
+   block and its ✓ / ✗ buttons, and the green/blue author colors, restyle to the
+   theme and stay legible in each.
+10. **Cleanliness.** `git status` shows nothing written to the document, sidecar,
+    or repo by the preview (INV-20).
+
+## Pass criteria
+
+All ten steps behave as described; no console errors in the webview devtools;
+the editor stays decoration-free throughout; nothing is persisted by the preview.
@@ -0,0 +1,67 @@
+# Manual smoke — F11 preview toolbar as the primary interaction surface (#43)
+
+F11 makes the rendered preview's **header toolbar** the primary interaction
+surface: beside the existing **Annotations** switch it gains a **Pin baseline**
+button and a **single adaptive Ask-Claude button** (Edit Selection when text is
+selected in the preview, Edit Document otherwise). The webview's *visual*
+behavior — the buttons, the live label flip, and the selection→source mapping —
+is verified **here**, not in the automated host E2E (the webview is a sealed
+sandbox; the host's message→seam wiring is covered by the F11 E2E suite). Run
+once per change that touches F11. Live turns hit the SDK; use a trivial
+instruction to keep them quick.
+
+## Setup
+
+1. `npm run build`
+2. Launch the Extension Development Host (F5 in VS Code, or the Run panel) with
+   `sandbox/` open.
+3. Open a markdown document with a few paragraphs (e.g. copy
+   `test/e2e/fixtures/workspace/docs/preview.md`).
+
+## Steps
+
+1. **Right-click gateway (PUC-6).** Right-click the markdown editor **tab** (or
+   use the editor title bar) → **"Cowriting: Open Review Preview"**. The preview
+   opens beside the editor. (The palette entry and `Ctrl+Alt+R` still work too.)
+2. **Three header controls (PUC-1).** The header row shows, left to right: the
+   **☑ Annotations** checkbox (on), a **⌖ Pin baseline** button, and a **✦ Ask
+   Claude to Edit Document** button. They restyle with the theme.
+3. **Adaptive label — selection (PUC-2).** Select a paragraph **in the preview**
+   (drag across rendered text). The Ask-Claude button label flips live to **"✦
+   Ask Claude to Edit Selection"**. Click elsewhere to clear the selection → it
+   flips back to **"✦ Ask Claude to Edit Document"**.
+4. **Edit Selection (PUC-3).** With a paragraph selected in the preview, click
+   **Ask Claude to Edit Selection** → a host input box appears → type an
+   instruction (e.g. "tighten this") → submit. A **blue `cw-proposal` block**
+   appears **inline at that paragraph** with **✓ / ✗** (the selection mapped back
+   to its source block; the edit ran host-side). The editor does **not** change
+   (INV-10). Accept with **✓**: the replacement lands; the block clears.
+5. **Edit Document (PUC-4).** Clear the selection (button reads "Edit Document").
+   Click it → instruct (e.g. "fix any typos and tighten") → submit. Claude's
+   whole-document rewrite surfaces as **one or more** independent blue proposal
+   blocks (one per changed hunk), each independently **✓ / ✗**-able. Accept some,
+   reject others — each behaves as an ordinary F10 proposal.
+6. **Pin baseline (PUC-5).** Make (or accept) a few changes so the preview shows
+   change-marks and a non-zero summary. Click **⌖ Pin baseline**. Expect: the
+   change-marks **clear**, the summary resets to `+0 −0`, and the header epoch
+   reads **`pinned <time>`**. "What changed" now counts from this moment.
+7. **Off-state still maps (INV-36).** Toggle **Annotations** off (clean preview).
+   Selecting a paragraph still flips the button to **Edit Selection** and an edit
+   still works — the off/clean preview is also a selection→source surface.
+8. **Non-authorable doc (PUC-1/7).** Open a markdown doc from a read-only/virtual
+   source (e.g. a Git diff view, or an Output channel rendered as markdown) and
+   open the preview. Expect: **Pin baseline** and **Ask-Claude** render
+   **disabled**; the **Annotations** toggle stays active (reading is allowed).
+9. **Theme.** Toggle light / dark / high-contrast (`Ctrl+K Ctrl+T`). The two
+   toolbar buttons (enabled, hover, disabled states) restyle and stay legible.
+10. **Cleanliness.** `git status` shows nothing written to the document, sidecar,
+    or repo by the toolbar gestures except the edits you explicitly accepted
+    (INV-20/35).
+
+## Pass criteria
+
+All ten steps behave as described; the adaptive label tracks the live selection;
+the selection-scoped edit anchors at the right paragraph; the document edit
+produces per-hunk proposals; pin clears the marks; edit controls are inert on a
+non-authorable doc; no console errors in the webview devtools; nothing is
+persisted by the toolbar beyond the edits you accepted.
@@ -0,0 +1,96 @@
+# Manual smoke — F12 document-edit flow
+
+Covers the document-edit-flow cluster (`specs/coauthoring-document-edit-flow.md`,
+#42 · #47 · #46). This file is filled in slice by slice.
+
+## SLICE-1 — #42 (reach): selection-aware Ask-Claude from body + tab (INV-38)
+
+Run the extension (F5) on a markdown document under the sandbox workspace.
+
+1. **Body, with selection (PUC-2).** Select a paragraph, right-click the editor
+   **body**. Expect **Ask Claude to Edit Selection** in the menu (and **not**
+   "Edit Document"). Pick it → instruct → submit; a single proposal lands over
+   the selection (existing F11 behavior, unchanged).
+2. **Body, no selection (PUC-1).** Clear the selection (click once), right-click
+   the editor **body**. Expect **Ask Claude to Edit Document** (and **not** "Edit
+   Selection"). Pick it → instruct → submit; the whole-document rewrite surfaces
+   as F4 proposal(s) in the preview.
+3. **Tab, with selection (PUC-3).** With a selection active, right-click the
+   editor **tab**. Expect **Ask Claude to Edit Selection**, acting on that tab's
+   document.
+4. **Tab, no selection (PUC-3).** With no selection, right-click the editor
+   **tab**. Expect **Ask Claude to Edit Document**, acting on **that tab's**
+   document — even if a *different* editor is the active one. Open two markdown
+   tabs A and B, make A active, right-click B's tab → Edit Document → the
+   proposals land on **B**, not A.
+5. **Markdown-gated.** Open a non-markdown file (e.g. `.txt`). Right-click body or
+   tab: neither **Ask Claude to Edit Selection** nor **Edit Document** appears.
+6. **Single edit path.** Both entries route through the same `runEditAndPropose`
+   path — there is no second edit code path (INV-38). Nothing is written to the
+   document or sidecar by merely invoking the menu (INV-10/20/35) until you accept.
+
+### Pass criteria
+
+The body and tab menus show exactly one Ask-Claude edit entry, matching the live
+selection state (selection → Edit Selection; none → Edit Document); the tab
+gesture targets the clicked tab's document, not the active editor; both are absent
+on non-markdown docs; no console errors.
+
+## SLICE-2 — #47 (review): per-block proposals + word-precise attribution (INV-39/40/41)
+
+On a markdown doc with several paragraphs, **Ask Claude to Edit Document** with a
+light copy-edit instruction (e.g. "tighten the prose, fix typos").
+
+1. **One proposal per changed block (INV-39).** Claude's pass surfaces as **one
+   ✓/✗ block per changed paragraph/header/bullet**, not a flurry of word-level
+   blocks. A paragraph with several word edits is a **single** proposal; the
+   word-level `<ins>`/`<del>` still shows *inside* it. Untouched paragraphs show no
+   proposal.
+2. **Changed fence is atomic (INV-23).** If Claude edits a code/mermaid fence, it
+   is **one** whole-fence proposal.
+3. **Accept attributes only the changed words (INV-40).** Accept a block proposal,
+   then toggle the preview to **Authorship**/colors (or re-open in the on-state):
+   only the words Claude actually changed are Claude-colored; the unchanged words
+   in that block keep their prior author. The block is the decision unit; the word
+   is the attribution unit.
+4. **Inserted block accepts cleanly (INV-41).** If Claude adds a new paragraph,
+   its proposal **accepts** (it is anchored to an adjacent block, never a
+   born-orphaned/zero-width proposal).
+5. **Undo.** Accepting a block is currently **N undo steps** (one per changed run
+   inside the block) — `Ctrl+Z` repeatedly restores it. (Spec deferred note:
+   single-undo-step grouping is a possible follow-up.)
+6. **Selection edits unchanged.** Edit *Selection* still produces exactly one
+   proposal over the selection (no block fan-out).
+
+### Pass criteria
+
+A document edit yields one in-context proposal per changed block (fences atomic);
+accepting a block lands the whole block but Claude-attributes only the words it
+changed; inserted blocks accept; selection edits are unaffected; no console errors.
+
+## SLICE-3 — #46 (accept): Accept all (INV-42)
+
+On a markdown doc, **Ask Claude to Edit Document** with a pass that changes
+**several** blocks, so the preview shows **≥ 2** pending proposals.
+
+1. **Button appears at ≥2 (PUC-6).** The preview toolbar shows **✓✓ Accept all**
+   only when there are **2 or more** pending proposals (and the doc is authorable,
+   annotations on). With 0–1 pending it is hidden.
+2. **One gesture applies all.** Click **Accept all**: every pending proposal lands
+   (Claude-attributed per INV-40), the ✓/✗ blocks clear, and a status message
+   reports how many were accepted. No confirmation dialog.
+3. **Undo restores.** `Ctrl+Z` walks back the applied edits (consistent with
+   single accept; a block accept is itself N steps — see SLICE-2).
+4. **Orphan-skip + report.** If one proposal's target text was changed by hand
+   first (so it can't anchor), Accept all applies the rest and the report says
+   `… , N skipped (target text changed — undo or reject)`; the orphaned proposal
+   stays pending, its text untouched (never force-applied).
+5. **Command path.** With no preview panel open, the command palette **Cowriting:
+   Accept All Claude Proposals** (markdown-gated) applies all proposals on the
+   active doc with the same report.
+
+### Pass criteria
+
+Accept all is offered only at ≥2 pending; one click applies every resolvable
+proposal and reports the tally; orphans are skipped (not mangled) and remain
+pending; the palette command works on the active doc; no console errors.
@@ -0,0 +1,28 @@
+# Manual smoke — F7.1 intra-diagram mermaid diffing (#22)
+
+Webview SVG rendering isn't covered by automated E2E (sealed sandbox, §6.8), so
+verify the rendered colors by hand once. The host-side augmentation (the styling
+directives injected into the mermaid source) *is* covered by unit + host E2E; this
+smoke confirms mermaid actually paints them.
+
+1. Open a markdown doc with a flowchart:
+
+   ```mermaid
+   flowchart LR
+     A[Start] --> B[Parse]
+     B --> C[Emit]
+   ```
+
+2. Run **Cowriting: Show Track-Changes Preview** (`ctrl+alt+r`). Pin the baseline
+   (**Cowriting: Pin Diff Baseline**).
+3. Edit the diagram: change `B[Parse]` → `B[Parse+Lint]`, add `B --> D[Validate]`,
+   and delete the `B --> C` edge and the `C[Emit]` node.
+4. **Expect** in the preview: `B` outlined amber (changed), `D` green (added),
+   `C` shown as a faded dashed ghost (removed) with a dashed grey ghost edge, and
+   a legend `added · changed · removed` beneath the diagram.
+5. Repeat with a `sequenceDiagram`: add a message (green `rect` band), remove one
+   (grey ghost band, message still visible), confirm a removed participant still
+   appears.
+6. Change a diagram to a `classDiagram` and confirm it still shows the v1
+   whole-block "changed" badge (graceful fallback, INV-30) — no error, no broken
+   render.
@@ -0,0 +1,38 @@
+# Manual smoke — F7 rendered track-changes preview (#21)
+
+The webview's *visual* rendering (mermaid, theming) is verified here, not in the
+automated host E2E (the webview is a sealed sandbox — §6.8). Run once per change
+that touches F7.
+
+## Setup
+
+1. `npm run build`
+2. Launch the Extension Development Host (F5 in VS Code, or the Run panel).
+3. Open a folder and a markdown document containing prose, a `mermaid` fenced
+   diagram, and a `ts` code fence (e.g. copy
+   `test/e2e/fixtures/workspace/docs/preview.md`).
+
+## Steps
+
+1. **Open the preview.** Run **"Cowriting: Open Track-Changes Preview"** (or
+   `Ctrl+Alt+R`). A preview opens beside the editor, rendering the document.
+   Header reads `Track changes since opened …`, summary `+0 −0`. The mermaid
+   diagram renders as a diagram; the code fence renders as highlighted code.
+2. **Edit prose.** In the source editor, change a word in a paragraph. The
+   preview updates (≈150 ms): the old word struck (`<del>`), the new word
+   highlighted (`<ins>`); the summary increments.
+3. **Edit the mermaid.** Change `a --> b` to `a --> c`. The preview re-renders the
+   **new** diagram with a **"changed"** badge at its top-right.
+4. **Ask Claude + accept.** Select the target sentence → "Ask Claude to Edit
+   Selection" → accept the proposal. The accepted text **drops its marks** (the
+   baseline advanced; PUC-3).
+5. **Pin.** Run "Cowriting: Pin Diff Baseline to Now". All marks clear (baseline
+   == now; PUC-4).
+6. **Theme.** Toggle light/dark (`Ctrl+K Ctrl+T`). Marks and the diagram restyle
+   to the theme.
+7. **Cleanliness.** `git status` shows nothing changed by the preview (INV-20).
+
+## Pass criteria
+
+All seven steps behave as described; no console errors in the webview devtools;
+nothing written to the document, sidecar, or repo.
@@ -0,0 +1,45 @@
+# Manual smoke — F8 out-of-workspace authoring
+
+Confirms "Ask Claude to Edit Selection" (+ threads / attribution) work on a file
+**outside** the workspace folder and on an **untitled** buffer, while in-workspace
+authoring is unchanged. One live turn hits the SDK (the only LLM step).
+
+Spec: `vscode-cowriting-plugin-content/specs/coauthoring-out-of-workspace.md`.
+
+## Prereqs
+- Build: `npm run build`.
+- Launch the Extension Development Host (F5 in VS Code) with this repo's `sandbox/`
+  opened as the workspace folder.
+
+## A. Out-of-folder file (PUC-1)
+1. Open a markdown file from a DIFFERENT directory (outside `sandbox/`) — e.g. a
+   sibling repo, or `File > Open` a scratch file under `/tmp`.
+2. Select a sentence → Command Palette → **Cowriting: Ask Claude to Edit Selection**
+   → type an instruction → wait for the proposal (amber range + ✓/✗).
+3. Accept (✓). Expect: the text is replaced and shows the Claude attribution tint.
+4. Add a coauthoring thread on a selection (**Add Coauthoring Thread on Selection**).
+5. Close and reopen the file (or revert). Expect: the thread + attribution are
+   restored. There is **no** `.threads/` folder beside the file — its state lives
+   in VS Code global storage (`<globalStorage>/sidecars/<hash>.json`).
+
+## B. Untitled buffer (PUC-2)
+1. `File > New File` (don't save) → type a few sentences.
+2. Select → **Ask Claude to Edit Selection** → instruct → accept. Expect: it works
+   exactly like A, in-session.
+3. Reload the window (Developer: Reload Window). Expect: the untitled buffer's
+   coauthoring state is **gone** (in-memory only — documented limitation, §6.7).
+
+## C. In-workspace unchanged (PUC-3)
+1. Open a file UNDER `sandbox/`. Repeat the propose→accept→thread loop.
+2. Expect: a committable `sandbox/.threads/<path>.json` sidecar is written, exactly
+   as before F8 (byte-for-byte, INV-2).
+
+## D. Read-only view declines (PUC-5)
+1. Open a Git diff / Output view, select text, run **Ask Claude to Edit Selection**.
+2. Expect: a warning that this kind of document can't be edited (not "select some
+   text").
+
+## E. Folder-less (optional)
+1. Launch the EDH with **no folder** open. Open an untitled buffer or `File > Open`
+   any file. Repeat A/B. Expect: authoring works (every doc routes to global
+   storage); the commands are not stubbed.
@@ -0,0 +1,21 @@
+# Manual smoke — F9 authorship view in the preview
+
+Confirms the rendered preview's Authorship mode colors Claude's vs your text. One
+live turn hits the SDK.
+
+## Prereqs
+- `npm run build`; launch the Extension Development Host (F5) with `sandbox/` open.
+
+## Steps
+1. Open a markdown file. Open **Cowriting: Open Track-Changes Preview** (`Ctrl+Alt+R`).
+2. The header shows a `[ Track changes | Authorship ]` toggle. It opens in **Track
+   changes** (unchanged behavior).
+3. Select a sentence → **Ask Claude to Edit Selection** → instruct → accept.
+4. Click **Authorship**. Expect: Claude's accepted text is tinted (blue) and your
+   own typing is tinted (green); the header shows a legend (Claude / You). Text you
+   never touched (original content) is plain.
+5. Type a few words yourself, mixing into a Claude sentence. Expect: the colors
+   split mid-paragraph at the exact boundaries.
+6. If the doc has a mermaid or code fence Claude authored, expect the whole fence
+   to carry a small author badge (not per-character).
+7. Flip back to **Track changes**. Expect: the diff view is exactly as before.
@@ -0,0 +1,995 @@
+# F10 — Interactive Track-Changes Review in the Markdown Preview — Implementation Plan
+
+> **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.
+
+**Goal:** Make the rendered markdown preview the single interactive review surface — a clean (zero-annotation) editor, an annotations on/off toggle, one combined per-author + strikethrough render, and ✓/✗ accept/reject of Claude's pending F4 proposals from inside the preview.
+
+**Architecture:** Mostly assembly of shipped features. F3 attribution (`spansFor`), F4 propose/accept (`acceptById`/`rejectById`), and the F6 baseline (`DiffViewController`) stay as **data layers**; their in-editor visuals are removed. One new pure, vscode-free function `renderReview(baselineText, currentText, authorSpans, proposals)` overlays the F7 block/word diff (author-colored via the salvaged F9 PUA-sentinel technique, struck deletions) **and** F4 pending proposals (blue blocks with ✓/✗). The `TrackChangesPreviewController` collapses F9's two modes into one `"on"|"off"` toggle, gains a `ProposalController` dependency, and routes ✓/✗ webview messages through the F4 seam. The webview never mutates the document (INV-20/21/34).
+
+**Tech Stack:** TypeScript, VS Code extension API, `markdown-it` + `diff` (jsdiff) + `mermaid` (webview-only), esbuild (webview bundle), vitest (vscode-free unit), `@vscode/test-electron` (host E2E).
+
+**Spec:** `specs/coauthoring-interactive-review.md` (F10, `#29`). Parent invariants INV-1..31 carry over except where F10 supersedes (F10 adds INV-32/33/34; **reverses F9 INV-26**).
+
+---
+
+## File Structure
+
+**Modify:**
+- `src/attributionController.ts` — `render()` stops applying editor decorations (keep `spansFor`); retire decoration types.
+- `src/proposalController.ts` — stop creating in-editor comment threads + pending decorations; refactor bookkeeping to plain maps; add `listProposals()` + `onDidChangeProposals`.
+- `src/trackChangesModel.ts` — add `renderReview` + `renderPlain` + `colorByAuthor` + `ProposalView`; remove public `renderAuthorship`.
+- `src/trackChangesPreview.ts` — collapse mode to `"on"|"off"`; take `ProposalController`; subscribe to `onDidChangeProposals`; new render path; inbound `accept`/`reject`/`setMode`; own a status-bar item.
+- `src/extension.ts` — reorder construction (proposals before preview); inject `ProposalController` into the preview.
+- `media/preview.ts` — replace segmented control with on/off switch; add ✓/✗ click → `postMessage`.
+- `media/preview.css` — proposal-block + ✓/✗ button styles; on/off switch style.
+- `package.json` — hide `toggleDiffView`/`ctrl+alt+d`, `toggleAttribution`, in-editor `acceptProposal`/`rejectProposal` (`when:false`); retitle `showTrackChangesPreview`.
+
+**Create:**
+- `docs/MANUAL-SMOKE-F10.md` — the live webview smoke checklist.
+
+**Test:**
+- `test/trackChangesModel.test.ts` — extend (vitest) for `renderReview`/`renderPlain`/`colorByAuthor`.
+- `test/e2e/suite/trackChangesPreview.test.ts` and/or a new `f10Review.test.ts` — host E2E.
+
+---
+
+## Conventions for this plan
+
+- **Unit tests** (the pure `trackChangesModel` engine) are strict TDD: failing test first, run it, minimal impl, run green, commit. Command: `npm test` (vitest) — to run one file: `npx vitest run test/trackChangesModel.test.ts`.
+- **vscode-layer changes** (controllers, webview, package.json) cannot be unit-tested vscode-free; they are verified by **host E2E** (`npm run test:e2e`) and the manual smoke. For those tasks the "test" step is the E2E assertion added in SLICE-4 plus a `npm run build` typecheck.
+- Build/typecheck after vscode-layer edits: `npm run build` (esbuild + bundles `media/preview.ts`). E2E precompiles via `npm run pretest:e2e`.
+- Commit after each task.
+
+---
+
+## SLICE-1 — Clean editor (INV-32)
+
+Strip F3 attribution decorations and F4 in-editor proposal threads/decorations; hide the F6 two-pane diff command/keybinding and the attribution toggle. Keep `spansFor`, accept/reject logic, and the baseline store. *Mostly deletion.*
+
+### Task 1: Strip F3 attribution editor decorations
+
+**Files:**
+- Modify: `src/attributionController.ts:62-63` (decoration types), `:354-373` (`render()`), `:82` (dispose list)
+
+- [ ] **Step 1: Read the current `render()` and decoration setup.** Confirm `render()` computes `agentRanges`/`humanRanges` and calls `editor.setDecorations(this.agentType, …)` / `(this.humanType, …)` at lines 366-367, and that `spansFor` (≈line 410) is independent of these.
+
+- [ ] **Step 2: Remove the editor-painting side effect from `render()`.** Edit `render()` so it no longer calls `setDecorations` and no longer builds `agentRanges`/`humanRanges`. F10 keeps `render()` as a method (callers exist) but it becomes a no-op for decorations — leave any non-decoration bookkeeping intact. Concretely, delete the two `editor.setDecorations(...)` calls (lines 366-367) and the range-building that feeds only them.
+
+- [ ] **Step 3: Retire the decoration types.** Delete the `agentType`/`humanType` `createTextEditorDecorationType(...)` fields (lines 62-63), the `AGENT_DECO`/`HUMAN_DECO` constants (≈lines 47-56), and remove `this.agentType, this.humanType` from the dispose push at line 82. Keep `this.statusItem, this.output, this.applyEmitter` in the dispose list.
+
+- [ ] **Step 4: Typecheck.**
+
+Run: `npm run build`
+Expected: no TypeScript errors; no remaining references to `agentType`/`humanType`/`AGENT_DECO`/`HUMAN_DECO`.
+
+- [ ] **Step 5: Confirm `spansFor` is untouched.** Grep that `spansFor(` still exists and returns `AuthorSpan[]`.
+
+Run: `grep -n 'spansFor' src/attributionController.ts`
+Expected: the method signature is present and unchanged.
+
+- [ ] **Step 6: Commit.**
+
+```bash
+git add src/attributionController.ts
+git commit -m "F10 SLICE-1: strip F3 attribution editor decorations (keep spansFor) (#29)"
+```
+
+### Task 2: Strip F4 in-editor proposal threads + decorations; refactor bookkeeping
+
+The editor must carry **no** F4 visuals. Remove the `CommentController`, the per-proposal comment threads, and the amber `pendingType` decoration — but keep resolving each proposal's anchor into `state.live`/`state.unresolved` so `getRendered` (existing tests) and SLICE-3's `listProposals` still work. The in-editor `acceptThread`/`rejectThread` paths become dead and are removed; `acceptById`/`rejectById` (the public seams) are kept.
+
+**Files:**
+- Modify: `src/proposalController.ts` — `DocState` (`:32-41`), constructor (`:49-71`), `renderAll` (`:187-210`), `renderProposal` (`:243-274`), `renderDecorations` (`:276-285`), `byThread`/`acceptThread`/`rejectThread` (`:137-144`, `:304-314`), `getRendered` (`:323-339`).
+
+- [ ] **Step 1: Drop the CommentController + pending decoration + thread bookkeeping.**
+  - Remove `private readonly controller = vscode.comments.createCommentController(...)` (≈line 64) and `private readonly pendingType = ...createTextEditorDecorationType(PENDING_DECO)` (line 53) and the `PENDING_DECO` constant (lines 43-47).
+  - Remove `this.controller` and `this.pendingType` from the dispose push (line 65) — keep `this.statusItem`.
+  - Remove the `acceptProposal`/`rejectProposal` command registrations (lines 67-68) — these are the in-editor thread-menu commands; the seams move to the preview in SLICE-3.
+  - In `DocState` remove `vsThreads: Map<string, vscode.CommentThread>` (line 36); keep `live` and `unresolved`.
+
+- [ ] **Step 2: Replace `renderProposal` (thread+UI) with `recordProposal` (bookkeeping only).** New method records resolved offsets without any vscode UI:
+
+```ts
+/** Record a proposal's resolved offsets (no editor UI — F10 INV-32). */
+private recordProposal(proposal: Proposal, offsets: OffsetRange, pending: boolean): void {
+  this.docs.get(this.keyOf2(proposal))?.live; // placeholder removed below
+}
+```
+
+  …actually keep it simple and stateful via the passed `state`:
+
+```ts
+private recordProposal(state: DocState, proposal: Proposal, offsets: OffsetRange, pending: boolean): void {
+  state.live.set(proposal.id, offsets);
+  if (!pending) state.unresolved.add(proposal.id);
+}
+```
+
+- [ ] **Step 3: Rewrite `renderAll` to use `recordProposal` and drop thread/decoration disposal.** Keep the anchor resolve-or-flag loop; remove `for (const vsThread of state.vsThreads.values()) vsThread.dispose();` and `state.vsThreads.clear();` and the `this.renderDecorations(...)` call:
+
+```ts
+renderAll(document: vscode.TextDocument): void {
+  if (!this.isTracked(document)) return;
+  const docPath = this.keyOf(document);
+  const state = this.ensureState(document);
+  state.artifact = this.store.load(docPath) ?? emptyArtifact(docPath);
+  state.live.clear();
+  state.unresolved.clear();
+  const text = document.getText();
+  for (const proposal of state.artifact.proposals) {
+    const fp = state.artifact.anchors[proposal.anchorId]?.fingerprint;
+    const resolved = fp ? resolve(text, fp) : "orphaned";
+    if (resolved === "orphaned") {
+      const line = fp ? Math.min(fp.lineHint, Math.max(0, document.lineCount - 1)) : 0;
+      const off = document.offsetAt(new vscode.Position(line, 0));
+      this.recordProposal(state, proposal, { start: off, end: off }, false);
+    } else {
+      this.recordProposal(state, proposal, resolved, true);
+    }
+  }
+  this.renderStatus(state);
+}
+```
+
+- [ ] **Step 4: Remove `renderDecorations` and the thread-following branch in `onDidChange`.** Delete `renderDecorations` (lines 276-285). In `onDidChange` (lines 222-239) keep the `state.live` shift loop but remove the `vsThread.range = ...` update and the trailing `this.renderDecorations(...)` call:
+
+```ts
+private onDidChange(e: vscode.TextDocumentChangeEvent): void {
+  const state = this.docs.get(this.keyOf(e.document));
+  if (!state || state.live.size === 0) return;
+  for (const change of e.contentChanges) {
+    const edit = { start: change.rangeOffset, end: change.rangeOffset + change.rangeLength, newLength: change.text.length };
+    for (const [id, range] of state.live) state.live.set(id, shift(range, edit));
+  }
+}
+```
+
+- [ ] **Step 5: Remove `acceptThread`/`rejectThread`/`byThread`.** Delete those three methods (lines 137-144, 304-314). `acceptById`/`rejectById`/`byId` stay. `accept`/`reject` keep their `renderAll(document)` re-render calls (now thread-free).
+
+- [ ] **Step 6: Repoint `getRendered` at `live`/`unresolved`.** It read from `vsThreads`; read from `live` instead:
+
+```ts
+getRendered(docPath: string): RenderedProposal[] {
+  const state = this.docs.get(docPath);
+  if (!state) return [];
+  const out: RenderedProposal[] = [];
+  for (const [id, off] of state.live) {
+    const p = state.artifact.proposals.find((x) => x.id === id)!;
+    out.push({
+      id,
+      pending: !state.unresolved.has(id),
+      canReply: false,
+      turnId: p.turnId,
+      range: { start: off.start, end: off.end },
+    });
+  }
+  return out;
+}
+```
+
+- [ ] **Step 7: Typecheck.**
+
+Run: `npm run build`
+Expected: no errors; no remaining references to `vsThreads`, `pendingType`, `PENDING_DECO`, `this.controller`, `acceptThread`, `rejectThread`, `byThread`, `renderDecorations`, `renderProposal`.
+
+- [ ] **Step 8: Run the existing F4 unit tests (proposalModel — vscode-free) to confirm no regression.**
+
+Run: `npx vitest run test/proposalModel.test.ts`
+Expected: PASS (proposalModel is untouched; this is a guard).
+
+- [ ] **Step 9: Commit.**
+
+```bash
+git add src/proposalController.ts
+git commit -m "F10 SLICE-1: remove F4 in-editor proposal threads/decorations; bookkeeping via live/unresolved (#29)"
+```
+
+> NOTE: the existing host E2E `test/e2e/suite/proposals.test.ts` may assert in-editor thread/`getRendered` behavior. Do **not** fix it here — SLICE-4 updates the E2E suite. If `npm run test:e2e` is run before SLICE-4, expect known proposal-thread failures.
+
+### Task 3: Hide F6 two-pane diff + attribution toggle in package.json
+
+**Files:**
+- Modify: `package.json` — keybindings (`:154-165`), `contributes.menus.commandPalette` (`:93-152`), command titles (`:48-91`)
+
+- [ ] **Step 1: Hide the `ctrl+alt+d` keybinding.** In `contributes.keybindings`, set the `toggleDiffView` entry's `when` to `false`:
+
+```json
+{ "command": "cowriting.toggleDiffView", "key": "ctrl+alt+d", "when": "false" }
+```
+
+- [ ] **Step 2: Hide `toggleDiffView`, `pinDiffBaseline`, and `toggleAttribution` from the command palette.** In `contributes.menus.commandPalette`, add (or set) `when: "false"` entries:
+
+```json
+{ "command": "cowriting.toggleDiffView", "when": "false" },
+{ "command": "cowriting.pinDiffBaseline", "when": "false" },
+{ "command": "cowriting.toggleAttribution", "when": "false" }
+```
+
+- [ ] **Step 3: Confirm the in-editor `acceptProposal`/`rejectProposal` palette entries are already `when:false`** (lines 104-110) and **remove their `editor/context`/`comments/commentThread` menu contributions** that referenced the now-deleted comment controller (the `commentController == cowriting.proposals` menu items, lines ≈143-151), since SLICE-1 deleted that controller.
+
+- [ ] **Step 4: Retitle `showTrackChangesPreview` and rebind to the review verb.** Set its title to `"Cowriting: Open Review Preview"` (line 89) and keep `ctrl+alt+r` / `when: editorLangId == markdown` (lines 161-164).
+
+- [ ] **Step 5: Validate JSON + typecheck.**
+
+Run: `node -e "JSON.parse(require('fs').readFileSync('package.json','utf8')); console.log('ok')" && npm run build`
+Expected: `ok` then a clean build.
+
+- [ ] **Step 6: Commit.**
+
+```bash
+git add package.json
+git commit -m "F10 SLICE-1: hide F6 diff command/keybinding + attribution toggle; retitle preview to 'Open Review Preview' (#29)"
+```
+
+---
+
+## SLICE-2 — Combined render engine (pure, vscode-free; INV-33)
+
+Add `renderReview` (the on-state) + `renderPlain` (the off-state) to `trackChangesModel.ts`, extracting the F9 PUA-sentinel coloring into a reusable `colorByAuthor`, and removing the now-superseded public `renderAuthorship`. Strict TDD — this is the pure core.
+
+### Task 4: Extract `colorByAuthor` from `renderAuthorship`'s inline sentinels
+
+**Files:**
+- Modify: `src/trackChangesModel.ts:326-352` (sentinel helpers), `:360-386` (`renderAuthorship`)
+- Test: `test/trackChangesModel.test.ts`
+
+- [ ] **Step 1: Write a failing unit test for `colorByAuthor`.** It should color a prose block's HTML by author spans (clipped to the block) using the existing sentinel technique.
+
+```ts
+import { colorByAuthor, type AuthorSpan } from "../src/trackChangesModel";
+
+test("colorByAuthor wraps human-authored prose in cw-by-human spans", () => {
+  const raw = "hello world";
+  const spans: AuthorSpan[] = [{ start: 0, end: 5, author: "human" }];
+  const render = (src: string) => `<p>${src}</p>`;
+  const html = colorByAuthor(raw, 0, spans, render);
+  expect(html).toContain('<span class="cw-by-human">hello</span>');
+  expect(html).toContain("world");
+});
+```
+
+- [ ] **Step 2: Run it — expect failure (`colorByAuthor` not exported).**
+
+Run: `npx vitest run test/trackChangesModel.test.ts -t colorByAuthor`
+Expected: FAIL — `colorByAuthor is not a function` / not exported.
+
+- [ ] **Step 3: Implement `colorByAuthor` by lifting the sentinel logic.** Export a function that injects sentinels for the block's overlapping spans, runs the provided render, and maps sentinels → spans:
+
+```ts
+/**
+ * Color one prose block's HTML by F3 author spans (PUA-sentinel technique,
+ * salvaged from F9 renderAuthorship). `blockStart` is the block's char offset in
+ * the source so spans map correctly. Pure; deterministic.
+ */
+export function colorByAuthor(
+  raw: string,
+  blockStart: number,
+  spans: AuthorSpan[],
+  render: (src: string) => string,
+): string {
+  const overlapping = spans.filter((s) => s.end > blockStart && s.start < blockStart + raw.length);
+  const injected = injectSentinels(raw, blockStart, overlapping);
+  return sentinelsToSpans(render(injected));
+}
+```
+
+  Keep `injectSentinels`, `sentinelsToSpans`, `SENT`, `isCloseSentinel`, `authorBadge` as the private helpers they already are.
+
+- [ ] **Step 4: Run the test — expect PASS.**
+
+Run: `npx vitest run test/trackChangesModel.test.ts -t colorByAuthor`
+Expected: PASS.
+
+- [ ] **Step 5: Refactor `renderAuthorship` to call `colorByAuthor` (no behavior change yet — it's removed in Task 7).** In its prose branch replace the inline `sentinelsToSpans(safe(injectSentinels(...)))` with `colorByAuthor(b.raw, b.start, overlapping, safe)`.
+
+- [ ] **Step 6: Run the full model suite — expect PASS (no regression).**
+
+Run: `npx vitest run test/trackChangesModel.test.ts`
+Expected: PASS (all existing renderAuthorship/renderTrackChanges cases green).
+
+- [ ] **Step 7: Commit.**
+
+```bash
+git add src/trackChangesModel.ts test/trackChangesModel.test.ts
+git commit -m "F10 SLICE-2: extract colorByAuthor from renderAuthorship sentinels (#29)"
+```
+
+### Task 5: Add `renderPlain` (the off-state)
+
+**Files:**
+- Modify: `src/trackChangesModel.ts`
+- Test: `test/trackChangesModel.test.ts`
+
+- [ ] **Step 1: Write a failing test.**
+
+```ts
+import { renderPlain } from "../src/trackChangesModel";
+
+test("renderPlain renders current buffer as plain markdown (no marks)", () => {
+  const html = renderPlain("# Title\n\nhello");
+  expect(html).toContain("<h1>Title</h1>");
+  expect(html).toContain("hello");
+  expect(html).not.toContain("cw-");
+});
+```
+
+- [ ] **Step 2: Run it — expect FAIL (not exported).**
+
+Run: `npx vitest run test/trackChangesModel.test.ts -t renderPlain`
+Expected: FAIL.
+
+- [ ] **Step 3: Implement.**
+
+```ts
+/** Off-state body: the current buffer as plain markdown, no annotations (INV-33). */
+export function renderPlain(currentText: string, opts: RenderOptions = {}): string {
+  const render = opts.render ?? defaultRender;
+  try {
+    return render(currentText);
+  } catch (err) {
+    return chip(err instanceof Error ? err.message : String(err));
+  }
+}
+```
+
+- [ ] **Step 4: Run — expect PASS.**
+
+Run: `npx vitest run test/trackChangesModel.test.ts -t renderPlain`
+Expected: PASS.
+
+- [ ] **Step 5: Commit.**
+
+```bash
+git add src/trackChangesModel.ts test/trackChangesModel.test.ts
+git commit -m "F10 SLICE-2: add renderPlain off-state (#29)"
+```
+
+### Task 6: Add `ProposalView` + `renderReview` (the on-state)
+
+`renderReview` overlays two axes in one pass: (1) the F7 baseline block/word diff, with added/changed prose author-colored via `colorByAuthor` and deletions struck; (2) each pending proposal injected at its resolved anchor as a blue `cw-proposal` block carrying `data-proposal-id` and a ✓/✗ placeholder. Unanchored proposals render as trailing blocks (never dropped — INV-34).
+
+**Files:**
+- Modify: `src/trackChangesModel.ts`
+- Test: `test/trackChangesModel.test.ts`
+
+- [ ] **Step 1: Add the `ProposalView` type.** (Resolved offsets are computed by the controller; the pure engine receives them.)
+
+```ts
+export interface ProposalView {
+  id: string;
+  /** resolved offsets in currentText; null when the anchor did not resolve. */
+  anchorStart: number | null;
+  anchorEnd: number | null;
+  /** the text the proposal would replace (fp.text), for the struck "before". */
+  replaced: string;
+  /** the proposed replacement text. */
+  replacement: string;
+}
+```
+
+- [ ] **Step 2: Write failing tests covering the on-state contract.**
+
+```ts
+import { renderReview, type ProposalView, type AuthorSpan } from "../src/trackChangesModel";
+
+test("renderReview: human addition since baseline renders green <ins>", () => {
+  const html = renderReview("hello", "hello world",
+    [{ start: 6, end: 11, author: "human" }], []);
+  expect(html).toMatch(/<ins>[^<]*world[^<]*<\/ins>|cw-by-human/);
+});
+
+test("renderReview: deletion since baseline renders struck cw-del/<del>", () => {
+  const html = renderReview("hello world", "hello", [], []);
+  expect(html).toMatch(/<del>|cw-del/);
+});
+
+test("renderReview: a pending proposal renders a blue block with data-proposal-id and a ✓/✗ action placeholder", () => {
+  const proposals: ProposalView[] = [
+    { id: "p1", anchorStart: 0, anchorEnd: 5, replaced: "hello", replacement: "goodbye" },
+  ];
+  const html = renderReview("hello", "hello", [], proposals);
+  expect(html).toContain('class="cw-proposal"');
+  expect(html).toContain('data-proposal-id="p1"');
+  expect(html).toContain("cw-actions");
+  expect(html).toContain("goodbye");
+  expect(html).toMatch(/<del>[^<]*hello[^<]*<\/del>|cw-del/);
+});
+
+test("renderReview: an unresolved proposal renders as a trailing block (never dropped)", () => {
+  const proposals: ProposalView[] = [
+    { id: "p2", anchorStart: null, anchorEnd: null, replaced: "x", replacement: "y" },
+  ];
+  const html = renderReview("a", "a", [], proposals);
+  expect(html).toContain('data-proposal-id="p2"');
+  expect(html).toContain("cw-proposal-unanchored");
+});
+
+test("renderReview is deterministic (same inputs → identical HTML)", () => {
+  const a = renderReview("hello", "hello world", [{ start: 6, end: 11, author: "human" }], []);
+  const b = renderReview("hello", "hello world", [{ start: 6, end: 11, author: "human" }], []);
+  expect(a).toBe(b);
+});
+
+test("renderReview: an atomic mermaid change is diffed whole (no inner sentinels)", () => {
+  const base = "```mermaid\nflowchart LR\n  A --> B\n```";
+  const cur = "```mermaid\nflowchart LR\n  A --> C\n```";
+  const html = renderReview(base, cur, [], []);
+  expect(html).toContain("mermaid");
+  expect(html).not.toContain("cw-by-"); // fences stay atomic
+});
+```
+
+- [ ] **Step 3: Run — expect FAIL (renderReview not exported).**
+
+Run: `npx vitest run test/trackChangesModel.test.ts -t renderReview`
+Expected: FAIL.
+
+- [ ] **Step 4: Implement `renderReview`.** Reuse `diffBlocks` + `renderOp`, but author-color added/changed prose and append proposal blocks. The simplest correct composition: render the diff body via the existing `renderOp` path **augmented** so added/changed prose `<ins>` content is author-colored, then inject proposal blocks. Add an internal `renderReviewBlock` that colors prose and a `proposalBlockHtml` emitter:
+
+```ts
+function proposalBlockHtml(p: ProposalView, render: (src: string) => string): string {
+  const safe = (src: string): string => {
+    try { return render(src); } catch (err) { return chip(err instanceof Error ? err.message : String(err)); }
+  };
+  const unanchored = p.anchorStart === null ? " cw-proposal-unanchored" : "";
+  const before = p.replaced ? `<del class="cw-del">${safe(p.replaced)}</del>` : "";
+  const after = `<ins class="cw-add">${safe(p.replacement)}</ins>`;
+  const actions =
+    `<span class="cw-actions">` +
+    `<button class="cw-accept" data-action="accept">✓</button>` +
+    `<button class="cw-reject" data-action="reject">✗</button>` +
+    `</span>`;
+  return `<div class="cw-proposal${unanchored}" data-proposal-id="${p.id}">${actions}${before}${after}</div>`;
+}
+
+/**
+ * On-state body (INV-33): the F7 baseline diff — added/changed PROSE author-colored
+ * via colorByAuthor (F9 sentinels), deletions struck — overlaid with F4 pending
+ * proposals as blue cw-proposal blocks (✓/✗). One pass, pure, vscode-free.
+ * Proposals are injected at their resolved anchor's block; unresolved ones append
+ * as trailing cw-proposal-unanchored blocks (never dropped — INV-34).
+ */
+export function renderReview(
+  baselineText: string,
+  currentText: string,
+  authorSpans: AuthorSpan[],
+  proposals: ProposalView[],
+  opts: RenderOptions = {},
+): string {
+  const render = opts.render ?? defaultRender;
+  // 1) the diff body, with prose author-coloring layered onto current-side blocks.
+  const ranges = splitBlocksWithRanges(currentText);
+  const ops = diffBlocks(baselineText, currentText);
+  // Map current-side blocks (added/unchanged/changed) to their source ranges for coloring.
+  const colored = (raw: string): string => {
+    const blk = ranges.find((r) => r.raw === raw);
+    if (!blk) return render(raw);
+    return colorByAuthor(raw, blk.start, authorSpans, render);
+  };
+  const bodyParts = ops.map((op) => renderReviewOp(op, render, colored));
+  // 2) proposal blocks. Resolved proposals are appended after the block containing
+  //    their anchor; for v1 simplicity (and determinism) append all proposals in id
+  //    order, anchored ones first, then unanchored — each carries its own marker.
+  const anchored = proposals.filter((p) => p.anchorStart !== null);
+  const unanchored = proposals.filter((p) => p.anchorStart === null);
+  const proposalParts = [...anchored, ...unanchored].map((p) => proposalBlockHtml(p, render));
+  return [...bodyParts, ...proposalParts].join("\n");
+}
+```
+
+  Add `renderReviewOp` — a thin wrapper over the existing `renderOp` that swaps the prose renderer for the author-coloring one on `added`/`changed`(non-atomic)/`unchanged` ops:
+
+```ts
+function renderReviewOp(
+  op: BlockOp,
+  render: (src: string) => string,
+  colored: (raw: string) => string,
+): string {
+  // Atomic fences and deletions: identical to renderOp (no author sentinels).
+  if (op.kind === "removed") return renderOp(op, render);
+  if (op.kind === "changed" && op.atomic) return renderOp(op, render);
+  // Prose added/unchanged: author-color the block; changed prose: keep <ins>/<del>
+  // word merge (deletions struck), then we accept that word-merge is not per-author
+  // colored in v1 (covered by §6.7 "deletion coloring = neutral").
+  if (op.kind === "changed") return renderOp(op, render); // word-merged <ins>/<del>
+  // unchanged / added prose → author-colored.
+  return `<div class="cw-blk ${op.kind === "added" ? "cw-added" : "cw-unchanged"}">${colored(op.block.raw)}</div>`;
+}
+```
+
+  > Design note: per spec §6.7, deletion coloring is neutral and added-prose author-coloring is the primary signal; changed-prose keeps the F7 word-merge. This keeps the engine deterministic and the tests above green. If a later refinement wants author-colored `<ins>` inside changed prose, it extends `renderReviewOp`.
+
+- [ ] **Step 5: Run the renderReview tests — expect PASS.**
+
+Run: `npx vitest run test/trackChangesModel.test.ts -t renderReview`
+Expected: PASS (all six).
+
+- [ ] **Step 6: Run the full model suite.**
+
+Run: `npx vitest run test/trackChangesModel.test.ts`
+Expected: PASS.
+
+- [ ] **Step 7: Commit.**
+
+```bash
+git add src/trackChangesModel.ts test/trackChangesModel.test.ts
+git commit -m "F10 SLICE-2: add ProposalView + renderReview combined on-state render (#29)"
+```
+
+### Task 7: Remove the superseded public `renderAuthorship`
+
+**Files:**
+- Modify: `src/trackChangesModel.ts` (remove `renderAuthorship`), `src/trackChangesPreview.ts` (stops importing it — done in SLICE-3, so here just remove the export and fix the model)
+- Test: `test/trackChangesModel.test.ts` (drop/replace renderAuthorship-only cases salvaged into colorByAuthor)
+
+- [ ] **Step 1: Delete the exported `renderAuthorship` function** (lines 360-386). Keep `colorByAuthor`, `injectSentinels`, `sentinelsToSpans`, `authorBadge`, `SENT`.
+
+- [ ] **Step 2: Remove or repoint any test that imports `renderAuthorship`.** Any remaining authorship-only assertions are now covered by the `colorByAuthor` test (Task 4); delete the obsolete `renderAuthorship` test cases.
+
+- [ ] **Step 3: Typecheck — expect a known error in `trackChangesPreview.ts`** (it still imports `renderAuthorship`). That import is removed in SLICE-3 Task 9; note it and proceed (do not patch the preview here beyond removing the dangling import if convenient).
+
+Run: `npm run build`
+Expected: error only at `trackChangesPreview.ts` `renderAuthorship` import → resolved in Task 9. (If you prefer a clean build per-task, remove the import + authorship branch now as a head-start on Task 9.)
+
+- [ ] **Step 4: Run the model suite — expect PASS.**
+
+Run: `npx vitest run test/trackChangesModel.test.ts`
+Expected: PASS.
+
+- [ ] **Step 5: Commit.**
+
+```bash
+git add src/trackChangesModel.ts test/trackChangesModel.test.ts
+git commit -m "F10 SLICE-2: remove superseded public renderAuthorship (salvaged into colorByAuthor) (#29)"
+```
+
+---
+
+## SLICE-3 — Interactive controller + webview (INV-34)
+
+Add `ProposalController.listProposals` + `onDidChangeProposals`; wire `ProposalController` into the preview; collapse mode to `"on"|"off"`; route ✓/✗/setMode messages; rebuild the webview asset; status-bar indicator.
+
+### Task 8: `ProposalController.listProposals` + `onDidChangeProposals`
+
+**Files:**
+- Modify: `src/proposalController.ts`
+
+- [ ] **Step 1: Add the change emitter + event.** Near the other fields:
+
+```ts
+private readonly onDidChangeProposalsEmitter = new vscode.EventEmitter<{ uri: string }>();
+/** Fires on propose / accept / reject / external sidecar change (F10). */
+readonly onDidChangeProposals = this.onDidChangeProposalsEmitter.event;
+```
+
+  Push it into `disposables` and fire it: at the end of `propose` (after `renderAll`), `accept` (after `removeProposal`+`renderAll`), `reject` (after `renderAll`), and `handleExternalSidecarChange`/`renderAll`. Fire with the document's URI string:
+
+```ts
+private fireChanged(document: vscode.TextDocument): void {
+  this.onDidChangeProposalsEmitter.fire({ uri: document.uri.toString() });
+}
+```
+
+  Call `this.fireChanged(document)` at the end of `renderAll(document)` (it is the common funnel for propose/accept/reject/external change).
+
+- [ ] **Step 2: Add `listProposals`.** Return resolved views for the preview's render. Resolve each proposal against the *current* document text (same as `renderAll`), exposing `fp.text` as `replaced`:
+
+```ts
+import type { ProposalView } from "./trackChangesModel";
+
+/** Resolved proposal views for the F10 preview (anchorStart=null when unresolved). */
+listProposals(document: vscode.TextDocument): ProposalView[] {
+  const docPath = this.keyOf(document);
+  const artifact = this.store.load(docPath) ?? emptyArtifact(docPath);
+  const text = document.getText();
+  return artifact.proposals.map((p) => {
+    const fp = artifact.anchors[p.anchorId]?.fingerprint;
+    const resolved = fp ? resolve(text, fp) : "orphaned";
+    return {
+      id: p.id,
+      anchorStart: resolved === "orphaned" ? null : resolved.start,
+      anchorEnd: resolved === "orphaned" ? null : resolved.end,
+      replaced: fp?.text ?? "",
+      replacement: p.replacement,
+    };
+  });
+}
+```
+
+- [ ] **Step 3: Typecheck.**
+
+Run: `npm run build`
+Expected: clean (the `ProposalView` import resolves against Task 6's export).
+
+- [ ] **Step 4: Commit.**
+
+```bash
+git add src/proposalController.ts
+git commit -m "F10 SLICE-3: ProposalController.listProposals + onDidChangeProposals (#29)"
+```
+
+### Task 9: Collapse preview mode to on/off, take `ProposalController`, route messages
+
+**Files:**
+- Modify: `src/trackChangesPreview.ts` (constructor, mode map, `refresh`, message handler, shell HTML, test seams, imports)
+
+- [ ] **Step 1: Update imports + mode type.** Replace the model import with the F10 surface and switch the mode union:
+
+```ts
+import { renderReview, renderPlain, diffBlocks, type BlockOp } from "./trackChangesModel";
+import type { ProposalController } from "./proposalController";
+```
+
+```ts
+/** F10: per-panel annotations toggle — on (combined render) or off (plain). */
+private readonly mode = new Map<string, "on" | "off">();
+```
+
+- [ ] **Step 2: Add the `ProposalController` constructor dependency + subscribe to its change event.**
+
+```ts
+constructor(
+  private readonly diffView: DiffViewController,
+  private readonly extensionUri: vscode.Uri,
+  private readonly attribution: AttributionController,
+  private readonly proposals: ProposalController,
+) {
+  this.disposables.push(
+    vscode.commands.registerCommand("cowriting.showTrackChangesPreview", () =>
+      this.show(vscode.window.activeTextEditor?.document),
+    ),
+    vscode.workspace.onDidChangeTextDocument((e) => this.onEdit(e.document)),
+    this.diffView.onDidChangeBaseline(({ uri }) => this.refreshByUri(uri)),
+    this.proposals.onDidChangeProposals(({ uri }) => { this.refreshByUri(uri); this.updateStatus(uri); }),
+  );
+}
+```
+
+- [ ] **Step 3: Replace the inbound message handler** (the `setMode` block) to also handle accept/reject:
+
+```ts
+panel.webview.onDidReceiveMessage(
+  (m: { type?: string; mode?: "on" | "off"; proposalId?: string }) => {
+    if (m?.type === "setMode" && (m.mode === "on" || m.mode === "off")) {
+      this.mode.set(key, m.mode);
+      this.refresh(document);
+    } else if (m?.type === "accept" && m.proposalId) {
+      void this.proposals
+        .acceptById(this.proposalKey(document), m.proposalId)
+        .then(() => this.refresh(document));
+    } else if (m?.type === "reject" && m.proposalId) {
+      this.proposals.rejectById(this.proposalKey(document), m.proposalId);
+      this.refresh(document);
+    }
+  },
+  null,
+  this.disposables,
+);
+```
+
+  Add a `proposalKey(document)` helper that returns the same key `ProposalController` uses (`this.store.keyOf(docIdentity(document))`). Since the preview doesn't hold the router, expose a public `keyFor(document): string` on `ProposalController` and call it:
+
+  In `proposalController.ts`:
+```ts
+/** The doc key F4 uses (F8 routing) — exposed for F10's preview. */
+keyFor(document: vscode.TextDocument): string { return this.keyOf(document); }
+```
+  In the preview, `private proposalKey(d: vscode.TextDocument) { return this.proposals.keyFor(d); }`.
+
+- [ ] **Step 4: Rewrite `refresh` for on/off.** Replace the `"authorship"`/`"changes"` branches with one combined `renderReview` (on) / `renderPlain` (off):
+
+```ts
+refresh(document: vscode.TextDocument): void {
+  const key = document.uri.toString();
+  const panel = this.panels.get(key);
+  if (!panel) return;
+  const mode = this.mode.get(key) ?? "on";
+  const current = document.getText();
+  const baseline = this.diffView.getBaseline(key);
+  const baselineText = baseline?.text ?? current; // no baseline → no change-marks
+  const ops = diffBlocks(baselineText, current);
+  this.lastModel.set(key, ops);
+  if (mode === "off") {
+    void panel.webview.postMessage({ type: "render", mode, html: renderPlain(current) });
+    return;
+  }
+  const spans = this.attribution.spansFor(document);
+  const proposals = this.proposals.listProposals(document);
+  const summary = {
+    added: ops.filter((o) => o.kind === "added").length,
+    removed: ops.filter((o) => o.kind === "removed").length,
+    proposals: proposals.length,
+  };
+  void panel.webview.postMessage({
+    type: "render",
+    mode,
+    html: renderReview(baselineText, current, spans, proposals),
+    epoch: this.epochLabel(baseline),
+    summary,
+  });
+}
+```
+
+- [ ] **Step 5: Update the shell HTML header** — replace the segmented `[Track changes | Authorship]` with an on/off switch:
+
+```ts
+<div id="cw-header">
+  <label id="cw-toggle"><input type="checkbox" id="cw-annotations" checked /> Annotations</label>
+  <span id="cw-epoch">Review</span>
+  <span id="cw-summary"></span>
+  <span id="cw-legend"></span>
+</div>
+```
+
+- [ ] **Step 6: Update the test seams** for the new mode type:
+
+```ts
+getMode(uriString: string): "on" | "off" { return this.mode.get(uriString) ?? "on"; }
+setMode(uriString: string, mode: "on" | "off"): void {
+  this.mode.set(uriString, mode);
+  const doc = vscode.workspace.textDocuments.find((d) => d.uri.toString() === uriString);
+  if (doc) this.refresh(doc);
+}
+/** F10 test seam: the on-state review HTML the panel would post. */
+renderHtmlFor(uriString: string): string {
+  const doc = vscode.workspace.textDocuments.find((d) => d.uri.toString() === uriString);
+  if (!doc) return "";
+  const current = doc.getText();
+  const baseline = this.diffView.getBaseline(uriString);
+  return renderReview(baseline?.text ?? current, current, this.attribution.spansFor(doc), this.proposals.listProposals(doc));
+}
+```
+
+- [ ] **Step 7: Add the status-bar item + `updateStatus`** (PUC-6) — see Task 10 (implement them here or in Task 10; cross-reference). For now stub `private updateStatus(_uri: string): void {}` so this task typechecks, and fill it in Task 10.
+
+- [ ] **Step 8: Typecheck.**
+
+Run: `npm run build`
+Expected: clean — no `renderAuthorship`/`renderTrackChanges` references remain in the preview.
+
+- [ ] **Step 9: Commit.**
+
+```bash
+git add src/trackChangesPreview.ts src/proposalController.ts
+git commit -m "F10 SLICE-3: preview takes ProposalController; on/off mode; renderReview path; accept/reject routing (#29)"
+```
+
+### Task 10: Status-bar indicator (PUC-6)
+
+A status-bar item shows the pending-proposal count when **no preview panel is open** for any doc with proposals; clicking it runs `cowriting.showTrackChangesPreview`. Hidden when count is 0 or a panel is open.
+
+**Files:**
+- Modify: `src/trackChangesPreview.ts`
+
+- [ ] **Step 1: Create the status-bar item in the constructor.**
+
+```ts
+private readonly statusItem = vscode.window.createStatusBarItem(vscode.StatusBarAlignment.Right, 88);
+```
+  In the constructor body: `this.statusItem.command = "cowriting.showTrackChangesPreview"; this.disposables.push(this.statusItem);`
+
+- [ ] **Step 2: Implement `updateStatus`.** Count pending proposals for the doc whose proposals changed; show the item only if that doc has no open panel:
+
+```ts
+private updateStatus(uri: string): void {
+  const doc = vscode.workspace.textDocuments.find((d) => d.uri.toString() === uri);
+  if (!doc) { this.statusItem.hide(); return; }
+  const n = this.proposals.listProposals(doc).length;
+  if (n === 0 || this.panels.has(uri)) { this.statusItem.hide(); return; }
+  this.statusItem.text = `$(comment-discussion) ${n} Claude proposal${n === 1 ? "" : "s"}`;
+  this.statusItem.tooltip = "Cowriting: open the review preview to accept/reject Claude's proposals";
+  this.statusItem.show();
+}
+```
+
+- [ ] **Step 3: Hide the status item when a panel opens.** In `show()` after `this.panels.set(key, panel)`, call `this.statusItem.hide()`. In `onDidDispose`, call `this.updateStatus(key)` so it reappears if proposals remain.
+
+- [ ] **Step 4: Dispose the status item.** Already pushed to `disposables`; confirm `dispose()` clears it.
+
+- [ ] **Step 5: Typecheck.**
+
+Run: `npm run build`
+Expected: clean.
+
+- [ ] **Step 6: Commit.**
+
+```bash
+git add src/trackChangesPreview.ts
+git commit -m "F10 SLICE-3: status-bar proposal indicator (PUC-6) (#29)"
+```
+
+### Task 11: Webview asset — on/off switch + ✓/✗ click handler + CSS
+
+**Files:**
+- Modify: `media/preview.ts`, `media/preview.css`
+
+- [ ] **Step 1: Update the inbound `RenderMessage` type + handler in `media/preview.ts`.** The mode is now `"on"|"off"`; `summary` carries `{added, removed, proposals}`:
+
+```ts
+type RenderMessage = {
+  type: "render";
+  mode: "on" | "off";
+  html: string;
+  epoch?: string;
+  summary?: { added: number; removed: number; proposals: number };
+};
+```
+
+  On receive: set `body.innerHTML = msg.html`; reflect the checkbox (`annotationsEl.checked = msg.mode === "on"`); show/hide summary/legend; then `renderMermaid()`.
+
+- [ ] **Step 2: Wire the on/off checkbox → `postMessage`.** Replace the `.cw-seg` segmented-control wiring:
+
+```ts
+const annotationsEl = document.getElementById("cw-annotations") as HTMLInputElement | null;
+annotationsEl?.addEventListener("change", () => {
+  vscode.postMessage({ type: "setMode", mode: annotationsEl.checked ? "on" : "off" });
+});
+```
+
+- [ ] **Step 3: Add the ✓/✗ delegated click handler.** Buttons are inside `.cw-proposal` blocks carrying `data-proposal-id`; read the id + `data-action` and post intent:
+
+```ts
+document.getElementById("cw-body")?.addEventListener("click", (e) => {
+  const btn = (e.target as HTMLElement)?.closest<HTMLElement>(".cw-actions button");
+  if (!btn) return;
+  const block = btn.closest<HTMLElement>(".cw-proposal");
+  const id = block?.dataset.proposalId;
+  const action = btn.dataset.action;
+  if (id && (action === "accept" || action === "reject")) {
+    vscode.postMessage({ type: action, proposalId: id });
+  }
+});
+```
+
+- [ ] **Step 4: Add CSS in `media/preview.css`** for the proposal block, ✓/✗ buttons, and on/off toggle, theme-variable-driven:
+
+```css
+.cw-proposal {
+  position: relative;
+  border-left: 3px solid var(--vscode-charts-blue, #4daafc);
+  background: color-mix(in srgb, var(--vscode-charts-blue, #4daafc) 12%, transparent);
+  padding: 0.4em 0.6em;
+  margin: 0.4em 0;
+  border-radius: 3px;
+}
+.cw-proposal-unanchored { border-left-style: dashed; opacity: 0.85; }
+.cw-actions { position: absolute; top: 0.2em; right: 0.4em; display: inline-flex; gap: 0.25em; }
+.cw-actions button {
+  cursor: pointer; border: 1px solid var(--vscode-button-border, transparent);
+  border-radius: 3px; font-size: 0.9em; line-height: 1; padding: 0.1em 0.35em;
+  background: var(--vscode-button-secondaryBackground); color: var(--vscode-button-secondaryForeground);
+}
+.cw-accept:hover { background: var(--vscode-testing-iconPassed, #2ea043); color: #fff; }
+.cw-reject:hover { background: var(--vscode-errorForeground, #f14c4c); color: #fff; }
+#cw-toggle { display: inline-flex; align-items: center; gap: 0.35em; cursor: pointer; }
+```
+
+- [ ] **Step 5: Build the webview bundle + typecheck.**
+
+Run: `npm run build`
+Expected: clean; `out/media/preview.js` rebuilt.
+
+- [ ] **Step 6: Commit.**
+
+```bash
+git add media/preview.ts media/preview.css
+git commit -m "F10 SLICE-3: webview on/off switch + ✓/✗ click→postMessage + proposal CSS (#29)"
+```
+
+### Task 12: Wire it together in `extension.ts`
+
+**Files:**
+- Modify: `src/extension.ts:92-105` (reorder: proposals before preview; inject)
+
+- [ ] **Step 1: Move `ProposalController` construction above `TrackChangesPreviewController`** and pass it in:
+
+```ts
+// --- F4: propose/accept (Feature #12) — constructed before the preview so F10
+// can route ✓/✗ through it ---
+const proposalController = new ProposalController(sidecarRouter, attributionController, root, versionGuard);
+context.subscriptions.push(proposalController);
+
+// --- F7/F10: the review preview is the single interactive review surface ---
+const trackChangesPreviewController = new TrackChangesPreviewController(
+  diffViewController,
+  context.extensionUri,
+  attributionController,
+  proposalController,
+);
+context.subscriptions.push(trackChangesPreviewController);
+```
+
+  Remove the now-duplicate later `proposalController` construction (old line 104).
+
+- [ ] **Step 2: Typecheck + full unit suite.**
+
+Run: `npm run build && npm test`
+Expected: clean build; vitest green (model + proposalModel + all vscode-free suites).
+
+- [ ] **Step 3: Commit.**
+
+```bash
+git add src/extension.ts
+git commit -m "F10 SLICE-3: wire ProposalController into the review preview (#29)"
+```
+
+---
+
+## SLICE-4 — Tests & docs
+
+### Task 13: Host E2E for the interactive review flow
+
+**Files:**
+- Modify: `test/e2e/suite/trackChangesPreview.test.ts` (or create `test/e2e/suite/f10Review.test.ts`), `test/e2e/suite/proposals.test.ts` (drop in-editor-thread assertions), `test/e2e/suite/authorship.test.ts` (remove/repoint F9 authorship-mode cases)
+
+- [ ] **Step 1: Repoint the obsolete F9 authorship E2E.** `authorship.test.ts` exercised `setMode("authorship")` + `renderAuthorship`. Either delete the file or rewrite its cases against `setMode(uri, "on")` and `renderReview` output (author-colored spans present). Confirm the suite index still imports valid files.
+
+- [ ] **Step 2: Fix `proposals.test.ts`.** Remove assertions about in-editor comment threads / `getRendered.canReply` semantics that no longer hold; keep `acceptById`/`rejectById` behavior assertions (those seams are unchanged).
+
+- [ ] **Step 3: Add the F10 host-E2E cases** (extend `trackChangesPreview.test.ts`). Use the existing harness patterns (open a markdown fixture, run `cowriting.showTrackChangesPreview`, drive the controller via its test seams and the `applyAgentEdit`/`proposeAgentEdit` commands). Assert:
+  - open markdown → panel open; `getLastModel` on a fresh baseline shows no change-marks.
+  - type text → an `added` BlockOp; the review HTML (`renderHtmlFor`) contains a `cw-by-human` span.
+  - `proposeAgentEdit` seam → `listProposals` returns one view with an id; `renderHtmlFor` contains `data-proposal-id` + `cw-actions`.
+  - simulate accept: call `proposalController.acceptById(key, id)` → proposal gone from `listProposals`, baseline advanced, the block now ordinary; document text reflects the replacement.
+  - simulate reject on a second proposal → gone from `listProposals`; document text unchanged.
+  - **clean editor:** assert no attribution decoration types are applied — e.g. assert `vscode.window.activeTextEditor` has no cowriting decorations (or that `AttributionController.render` no longer creates them; a structural assertion that the decoration-type fields were removed).
+  - `toggleDiffView` is hidden: assert its keybinding `when` is `false` (read from `package.json`) or that invoking it is a no-op user-facing.
+  - non-markdown doc → `show` warns, no panel (`isOpen` false).
+  - status-bar: after a propose with no panel open for that doc, the controller's status path reports a pending count (assert via a small seam if needed, e.g. expose `statusText()` for tests).
+
+- [ ] **Step 4: Add a tiny test seam if needed.** If asserting the status bar is awkward, add `statusText(): string | undefined` returning `this.statusItem.text` when visible.
+
+- [ ] **Step 5: Run the host E2E suite.**
+
+Run: `npm run test:e2e`
+Expected: PASS (all suites, including the updated proposals/authorship and new F10 cases).
+
+- [ ] **Step 6: Run the full unit suite too.**
+
+Run: `npm test`
+Expected: PASS.
+
+- [ ] **Step 7: Commit.**
+
+```bash
+git add test
+git commit -m "F10 SLICE-4: host E2E for interactive review (open/toggle/propose→accept→reject/clean-editor/hidden-F6/status) (#29)"
+```
+
+### Task 14: Manual smoke doc + README
+
+**Files:**
+- Create: `docs/MANUAL-SMOKE-F10.md`
+- Modify: `README.md` (add an F10 section; note F6 two-pane + F9 authorship-mode are superseded as user surfaces)
+
+- [ ] **Step 1: Write `docs/MANUAL-SMOKE-F10.md`** following the F7/F9 smoke format. Checklist: open a markdown doc → editor is clean (no tint/threads); edit prose → green `<ins>` / struck `<del>` in the preview; ask Claude to edit a selection → a blue proposal block with ✓/✗ appears; click ✓ (lands, mark clears, editor updated) and ✗ on another (block vanishes, doc unchanged); toggle Annotations off (clean render) / on; with no preview open, the status-bar indicator shows the pending count and opens the preview when clicked; verify light/dark/high-contrast theming; verify `git status` shows nothing persisted.
+
+- [ ] **Step 2: Update `README.md`.** Add an F10 entry to the feature list describing "write left / review right," the annotations on/off toggle, and ✓/✗ in the preview; add a one-line note that F6's two-pane diff and F9's authorship mode are retained as data layers but no longer separate user surfaces; mention `ctrl+alt+r` opens the review preview.
+
+- [ ] **Step 3: Commit.**
+
+```bash
+git add docs/MANUAL-SMOKE-F10.md README.md
+git commit -m "F10 SLICE-4: manual smoke checklist + README F10 section (#29)"
+```
+
+---
+
+## Final verification (before PR)
+
+- [ ] `npm test` — vitest green (model incl. renderReview/renderPlain/colorByAuthor; proposalModel; all vscode-free suites).
+- [ ] `npm run test:e2e` — host E2E green (open/toggle/propose→accept→reject/clean-editor/hidden-F6/status/non-markdown).
+- [ ] `npm run build` — clean typecheck + bundles.
+- [ ] Manual smoke per `docs/MANUAL-SMOKE-F10.md` performed once (the webview visual + real button clicks the E2E can't cover).
+- [ ] `git status` clean (nothing persisted by the preview — INV-20).
+- [ ] Open PR to `main`, citing `#29` and `specs/coauthoring-interactive-review.md`; acceptance per spec §7.3.
+
+---
+
+## Self-Review notes (spec coverage)
+
+- INV-32 (clean editor) → SLICE-1 (Tasks 1-3). INV-33 (combined pure render) → SLICE-2 (Tasks 4-7). INV-34 (✓/✗ via F4 seam) → SLICE-3 (Tasks 8-12).
+- PUC-1 clean editor → Task 1/2; PUC-2 toggle → Task 9/11; PUC-3 who-changed-what → Task 6; PUC-4/5 accept/reject → Task 9 (routing) + Task 6 (blocks); PUC-6 status-bar → Task 10; PUC-7 edges (non-markdown warn, no-baseline note, unanchored proposal) → Task 6 (unanchored), Task 9 (no-baseline), `show()` (non-markdown, existing).
+- F9 INV-26 reversal (authorship combined with diff; remove segmented control/`renderAuthorship`) → Tasks 7, 9, 13(step1).
+- Supersession: F6 two-pane hidden → Task 3; F9 INV-27/28 absorbed into renderReview → Task 6.
+- Reconciliation captured: public seams `acceptById`/`rejectById` (Task 9), `Proposal.anchorId` resolved via `resolve()` in `listProposals` (Task 8), new `onDidChangeProposals` (Task 8), `keyFor` exposure for the preview's doc key (Task 9).
@@ -0,0 +1,788 @@
+# F9 Authorship Preview Implementation Plan
+
+> **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.
+
+**Goal:** Add an **Authorship** mode to the F7 rendered preview — a header toggle that re-renders the current markdown doc with each span colored by its F3 author (Claude blue / human green), inline and char-precise — so a writer can see what Claude composed even after the F6 baseline absorbs it.
+
+**Architecture:** A new pure render engine `renderAuthorship(currentText, authorSpans)` in `src/trackChangesModel.ts` (vscode-free, unit-tested) injects Private-Use-Area sentinels at attribution-span boundaries through markdown-it, then post-processes them into `<span class="cw-by-claude|cw-by-human">`; code/mermaid fences stay atomic with a block-level author badge. `AttributionController.spansFor(document)` supplies the spans (resolving the URI-vs-keyOf key mismatch). `TrackChangesPreviewController` gains a per-panel mode and an attribution dependency; the webview header gets a segmented `[ Track changes | Authorship ]` toggle that posts `setMode` back.
+
+**Tech Stack:** TypeScript, VS Code webview API, `markdown-it`, vitest (unit), `@vscode/test-electron` + mocha (host E2E). Design: `docs/superpowers/specs/2026-06-11-authorship-preview-design.md`.
+
+**Conventions (read first):**
+- The render engine is **vscode-free** and **deterministic** (INV-22/28) — spans are passed as data, tested under vitest like `test/trackChangesModel.test.ts`.
+- Avoid `*/` inside block comments (an earlier F8 hiccup: a `**/` glob closed a comment early). Keep glob examples out of JSDoc.
+- Attribution offsets are char offsets into `document.getText()` — the same string passed to the renderer as `currentText`, so they align.
+- Run unit: `npm test`. Typecheck: `npm run typecheck`. Build: `npm run build`. Host E2E: `npm run test:e2e`.
+- After each task: `npm test` green + `npm run typecheck` clean, then commit.
+
+---
+
+## File Structure
+
+**Modify:**
+- `src/trackChangesModel.ts` — add `AuthorKind`/`AuthorSpan`, `splitBlocksWithRanges`, `renderAuthorship`.
+- `src/attributionController.ts` — add `spansFor(document): AuthorSpan[]`.
+- `src/trackChangesPreview.ts` — attribution dependency; per-panel mode; `setMode` handling; render branch; post `mode`/`legend`.
+- `src/extension.ts` — reorder so attribution exists before the F7 controller; pass it in.
+- `media/preview.ts` — segmented toggle + `acquireVsCodeApi` postMessage; legend vs summary per mode.
+- `media/preview.css` — `.cw-by-claude` / `.cw-by-human`, toggle, legend, fence-badge styles.
+- `README.md` — F9 note.
+
+**Create:**
+- `test/e2e/suite/authorship.test.ts` — host E2E.
+- `docs/MANUAL-SMOKE-F9.md` — manual smoke runbook.
+
+**Untouched:** `renderTrackChanges`/`diffBlocks` behavior (track-changes mode), F6 baseline, F3 capture, the seam, persistence.
+
+---
+
+## Task 1: `splitBlocksWithRanges` (block offsets)
+
+**Files:**
+- Modify: `src/trackChangesModel.ts`
+- Test: `test/trackChangesModel.test.ts`
+
+- [ ] **Step 1: Write the failing test (append to `test/trackChangesModel.test.ts`)**
+
+```typescript
+import { splitBlocksWithRanges } from "../src/trackChangesModel";
+
+describe("splitBlocksWithRanges — block offsets align with the source string", () => {
+  it("each block's [start,end) slices back to its raw from the source", () => {
+    const text = "# Title\n\nA prose paragraph.\n\n```js\ncode()\n```\n";
+    const blocks = splitBlocksWithRanges(text);
+    expect(blocks.map((b) => b.type)).toEqual(["prose", "prose", "code"]);
+    for (const b of blocks) {
+      expect(text.slice(b.start, b.end)).toBe(b.raw);
+    }
+    expect(blocks[1].raw).toBe("A prose paragraph.");
+  });
+
+  it("marks a mermaid fence and preserves its source offsets", () => {
+    const text = "intro\n\n```mermaid\ngraph TD; A-->B;\n```\n";
+    const blocks = splitBlocksWithRanges(text);
+    expect(blocks[1].type).toBe("mermaid");
+    expect(text.slice(blocks[1].start, blocks[1].end)).toBe(blocks[1].raw);
+  });
+});
+```
+
+- [ ] **Step 2: Run to verify it fails**
+
+Run: `npm test -- trackChangesModel`
+Expected: FAIL (`splitBlocksWithRanges` is not exported).
+
+- [ ] **Step 3: Implement `splitBlocksWithRanges` in `src/trackChangesModel.ts`**
+
+Add after `splitBlocks`:
+
+```typescript
+export interface BlockWithRange extends Block {
+  /** char offset of the block's first char in the source string. */
+  start: number;
+  /** char offset one past the block's last char (source.slice(start,end) === raw). */
+  end: number;
+}
+
+/**
+ * Like splitBlocks, but each block carries its [start,end) char range in the
+ * SOURCE string (offsets align with document.getText(), so F3 attribution
+ * offsets map directly). Lines are tracked with their true source offsets
+ * (a trailing CR stays in the line, so \r\n sources keep correct offsets).
+ */
+export function splitBlocksWithRanges(text: string): BlockWithRange[] {
+  // line raw (excluding the \n terminator) + its source start offset.
+  const lines: { raw: string; start: number }[] = [];
+  let from = 0;
+  for (let i = 0; i <= text.length; i++) {
+    if (i === text.length || text[i] === "\n") {
+      lines.push({ raw: text.slice(from, i), start: from });
+      from = i + 1;
+      if (i === text.length) break;
+    }
+  }
+  const out: BlockWithRange[] = [];
+  const range = (lo: number, hi: number): { start: number; end: number } => {
+    const start = lines[lo].start;
+    const end = lines[hi].start + lines[hi].raw.length;
+    return { start, end };
+  };
+  let buf: number[] = []; // indices of buffered prose lines
+  const flushProse = () => {
+    if (buf.length) {
+      const { start, end } = range(buf[0], buf[buf.length - 1]);
+      const raw = text.slice(start, end);
+      if (raw.trim()) out.push({ ...makeBlockRange(raw, "prose"), start, end });
+    }
+    buf = [];
+  };
+  let i = 0;
+  while (i < lines.length) {
+    const line = lines[i].raw;
+    const fence = line.match(/^(\s*)(`{3,}|~{3,})(.*)$/);
+    if (fence) {
+      flushProse();
+      const marker = fence[2][0];
+      const info = fence[3].trim().split(/\s+/)[0].toLowerCase();
+      const open = i;
+      i++;
+      while (i < lines.length) {
+        const closed = lines[i].raw.trim().startsWith(marker.repeat(3));
+        i++;
+        if (closed) break;
+      }
+      const close = i - 1;
+      const { start, end } = range(open, close);
+      const raw = text.slice(start, end);
+      out.push({ ...makeBlockRange(raw, info === "mermaid" ? "mermaid" : "code"), start, end });
+      continue;
+    }
+    if (line.trim() === "") {
+      flushProse();
+      i++;
+      continue;
+    }
+    buf.push(i);
+    i++;
+  }
+  flushProse();
+  return out;
+}
+
+function makeBlockRange(raw: string, type: BlockType): Block {
+  return makeBlock(raw, type);
+}
+```
+
+(`makeBlockRange` is a thin alias so the spread `{ ...makeBlockRange(...), start, end }` reads clearly; it reuses the existing `makeBlock`.)
+
+- [ ] **Step 4: Run to verify it passes**
+
+Run: `npm test -- trackChangesModel`
+Expected: PASS.
+
+- [ ] **Step 5: Commit**
+
+```bash
+git add src/trackChangesModel.ts test/trackChangesModel.test.ts
+git commit -m "feat(f9): splitBlocksWithRanges — block offsets aligned to the source"
+```
+
+---
+
+## Task 2: `renderAuthorship` — inline spans + atomic fences
+
+**Files:**
+- Modify: `src/trackChangesModel.ts`
+- Test: `test/trackChangesModel.test.ts`
+
+- [ ] **Step 1: Write the failing tests (append)**
+
+```typescript
+import { renderAuthorship, type AuthorSpan } from "../src/trackChangesModel";
+
+const spanAt = (text: string, sub: string, author: "claude" | "human"): AuthorSpan => {
+  const start = text.indexOf(sub);
+  return { start, end: start + sub.length, author };
+};
+
+describe("renderAuthorship", () => {
+  it("empty spans → plain render (no author wrappers)", () => {
+    const text = "# Hi\n\nplain paragraph.\n";
+    const html = renderAuthorship(text, []);
+    expect(html).not.toContain("cw-by-claude");
+    expect(html).not.toContain("cw-by-human");
+    expect(html).toContain("plain paragraph.");
+  });
+
+  it("wraps a single Claude span inline", () => {
+    const text = "The cat sat on the mat.\n";
+    const html = renderAuthorship(text, [spanAt(text, "cat sat", "claude")]);
+    expect(html).toContain('<span class="cw-by-claude">cat sat</span>');
+  });
+
+  it("marks two authors in one paragraph at exact boundaries", () => {
+    const text = "Alpha beta gamma.\n";
+    const html = renderAuthorship(text, [
+      spanAt(text, "Alpha", "human"),
+      spanAt(text, "gamma", "claude"),
+    ]);
+    expect(html).toContain('<span class="cw-by-human">Alpha</span>');
+    expect(html).toContain('<span class="cw-by-claude">gamma</span>');
+  });
+
+  it("clips a span to its block (does not bleed across blocks)", () => {
+    const text = "Para one.\n\nPara two.\n";
+    // a span covering the whole text; each block wraps only its own slice
+    const html = renderAuthorship(text, [{ start: 0, end: text.length, author: "claude" }]);
+    expect(html).toContain('<span class="cw-by-claude">Para one.</span>');
+    expect(html).toContain('<span class="cw-by-claude">Para two.</span>');
+  });
+
+  it("a code fence overlapping a span gets a block badge, NOT inner sentinels (atomic)", () => {
+    const text = "```js\nconst x = 1;\n```\n";
+    const html = renderAuthorship(text, [{ start: 0, end: text.length, author: "claude" }]);
+    expect(html).toContain("cw-by-claude");
+    expect(html).toContain("cw-badge");
+    expect(html).not.toContain(""); // no sentinel leaked into the code
+    expect(html).toContain("const x = 1;");
+  });
+
+  it("a mermaid fence authored by Claude renders as a diagram with a badge", () => {
+    const text = "```mermaid\ngraph TD; A-->B;\n```\n";
+    const html = renderAuthorship(text, [{ start: 0, end: text.length, author: "claude" }]);
+    expect(html).toContain('pre class="mermaid"');
+    expect(html).toContain("cw-by-claude");
+    expect(html).toContain("cw-badge");
+  });
+
+  it("is deterministic", () => {
+    const text = "Stable input paragraph.\n";
+    const spans: AuthorSpan[] = [spanAt(text, "input", "claude")];
+    expect(renderAuthorship(text, spans)).toBe(renderAuthorship(text, spans));
+  });
+});
+```
+
+- [ ] **Step 2: Run to verify it fails**
+
+Run: `npm test -- trackChangesModel`
+Expected: FAIL (`renderAuthorship`/`AuthorSpan` not exported).
+
+- [ ] **Step 3: Implement in `src/trackChangesModel.ts`**
+
+```typescript
+export type AuthorKind = "claude" | "human";
+export interface AuthorSpan {
+  start: number;
+  end: number;
+  author: AuthorKind;
+}
+
+// Private-Use-Area sentinels (never appear in real content; markdown-it passes
+// them through as plain text). Paired open/close per author.
+const SENT = {
+  claude: { open: "", close: "" },
+  human: { open: "", close: "" },
+} as const;
+
+function authorBadge(authors: Set<AuthorKind>): { cls: string; label: string } | null {
+  if (authors.size === 0) return null;
+  if (authors.size > 1) return { cls: "cw-mixed", label: "mixed" };
+  const only = [...authors][0];
+  return only === "claude"
+    ? { cls: "cw-by-claude", label: "Claude" }
+    : { cls: "cw-by-human", label: "You" };
+}
+
+/** Inject paired sentinels into a prose block's raw text for the spans clipped to it. */
+function injectSentinels(raw: string, blockStart: number, spans: AuthorSpan[]): string {
+  // Build insertions as (localOffset, marker); apply right-to-left so offsets stay valid.
+  const inserts: { at: number; marker: string }[] = [];
+  for (const s of spans) {
+    const lo = Math.max(0, s.start - blockStart);
+    const hi = Math.min(raw.length, s.end - blockStart);
+    if (hi <= lo) continue;
+    inserts.push({ at: lo, marker: SENT[s.author].open });
+    inserts.push({ at: hi, marker: SENT[s.author].close });
+  }
+  // Close markers must come BEFORE open markers at the same offset to keep nesting
+  // tidy; otherwise sort by offset descending and, at equal offset, close first.
+  inserts.sort((a, b) => (b.at - a.at) || (isClose(a.marker) ? -1 : 1));
+  let out = raw;
+  for (const ins of inserts) out = out.slice(0, ins.at) + ins.marker + out.slice(ins.at);
+  return out;
+}
+function isClose(m: string): boolean {
+  return m === SENT.claude.close || m === SENT.human.close;
+}
+
+/** Replace the rendered sentinels with author <span> tags. */
+function sentinelsToSpans(html: string): string {
+  return html
+    .split(SENT.claude.open).join('<span class="cw-by-claude">')
+    .split(SENT.claude.close).join("</span>")
+    .split(SENT.human.open).join('<span class="cw-by-human">')
+    .split(SENT.human.close).join("</span>");
+}
+
+/**
+ * Pure authorship render (INV-26/28): the CURRENT text with each F3-attributed
+ * span colored by author. Prose blocks get inline `<span class="cw-by-*">`;
+ * code/mermaid fences stay ATOMIC (INV-27) — an overlapping span yields a
+ * block-level author badge, never inner sentinels. Deterministic.
+ */
+export function renderAuthorship(
+  currentText: string,
+  spans: AuthorSpan[],
+  opts: RenderOptions = {},
+): string {
+  const render = opts.render ?? defaultRender;
+  const safe = (src: string): string => {
+    try {
+      return render(src);
+    } catch (err) {
+      return chip(err instanceof Error ? err.message : String(err));
+    }
+  };
+  return splitBlocksWithRanges(currentText)
+    .map((b) => {
+      const overlapping = spans.filter((s) => s.end > b.start && s.start < b.end);
+      if (b.type !== "prose") {
+        const badge = authorBadge(new Set(overlapping.map((s) => s.author)));
+        const inner = safe(b.raw);
+        if (!badge) return `<div class="cw-blk">${inner}</div>`;
+        return `<div class="cw-blk ${badge.cls}"><span class="cw-badge">${badge.label}</span>${inner}</div>`;
+      }
+      const injected = injectSentinels(b.raw, b.start, overlapping);
+      return `<div class="cw-blk">${sentinelsToSpans(safe(injected))}</div>`;
+    })
+    .join("\n");
+}
+```
+
+- [ ] **Step 4: Run to verify it passes**
+
+Run: `npm test -- trackChangesModel`
+Expected: PASS. If the inline-span test fails because markdown-it wraps prose in `<p>`, the assertion still holds (`<span ...>cat sat</span>` appears inside the `<p>`). If a sentinel survives in output, check `sentinelsToSpans` ordering.
+
+- [ ] **Step 5: Commit**
+
+```bash
+git add src/trackChangesModel.ts test/trackChangesModel.test.ts
+git commit -m "feat(f9): renderAuthorship — inline author spans + atomic fence badges (INV-26/27/28)"
+```
+
+---
+
+## Task 3: `AttributionController.spansFor`
+
+**Files:**
+- Modify: `src/attributionController.ts`
+
+- [ ] **Step 1: Add the method**
+
+In `src/attributionController.ts`, import the type and add a public method near `getSpans`:
+
+```typescript
+import type { AuthorSpan } from "./trackChangesModel";
+```
+
+```typescript
+  /**
+   * F9: the document's live attribution as authorship spans for the preview —
+   * current-buffer char ranges mapped to author kind (agent→claude). Computes
+   * the document key internally (so callers pass a TextDocument, not the key).
+   */
+  spansFor(document: vscode.TextDocument): AuthorSpan[] {
+    return this.getSpans(this.keyOf(document)).map((s) => ({
+      start: s.range.start,
+      end: s.range.end,
+      author: s.authorKind === "agent" ? "claude" : "human",
+    }));
+  }
+```
+
+- [ ] **Step 2: Verify typecheck**
+
+Run: `npm run typecheck`
+Expected: clean.
+
+- [ ] **Step 3: Commit**
+
+```bash
+git add src/attributionController.ts
+git commit -m "feat(f9): AttributionController.spansFor — authorship spans for the preview"
+```
+
+---
+
+## Task 4: Wire mode + attribution into the preview controller
+
+**Files:**
+- Modify: `src/trackChangesPreview.ts`, `src/extension.ts`
+
+- [ ] **Step 1: Controller — attribution dep, mode state, render branch, setMode**
+
+In `src/trackChangesPreview.ts`:
+
+Imports + fields:
+```typescript
+import { renderTrackChanges, renderAuthorship, diffBlocks, type BlockOp } from "./trackChangesModel";
+import type { AttributionController } from "./attributionController";
+```
+Add a per-panel mode map field:
+```typescript
+  private readonly mode = new Map<string, "changes" | "authorship">();
+```
+Constructor — add the attribution param (after `extensionUri`):
+```typescript
+  constructor(
+    private readonly diffView: DiffViewController,
+    private readonly extensionUri: vscode.Uri,
+    private readonly attribution: AttributionController,
+  ) {
+```
+
+In `show(...)`, after creating the panel, wire incoming messages (the toggle). Add to the `panel.onDidDispose` registration area:
+```typescript
+    panel.webview.onDidReceiveMessage(
+      (m: { type?: string; mode?: "changes" | "authorship" }) => {
+        if (m?.type === "setMode" && (m.mode === "changes" || m.mode === "authorship")) {
+          this.mode.set(key, m.mode);
+          this.refresh(document);
+        }
+      },
+      null,
+      this.disposables,
+    );
+```
+
+Rewrite `refresh(document)` to branch on mode:
+```typescript
+  refresh(document: vscode.TextDocument): void {
+    const key = document.uri.toString();
+    const panel = this.panels.get(key);
+    if (!panel) return;
+    const mode = this.mode.get(key) ?? "changes";
+    const current = document.getText();
+    if (mode === "authorship") {
+      const spans = this.attribution.spansFor(document);
+      void panel.webview.postMessage({
+        type: "render",
+        mode,
+        html: renderAuthorship(current, spans),
+        legend: { claude: spans.some((s) => s.author === "claude"), human: spans.some((s) => s.author === "human") },
+      });
+      this.lastModel.set(key, diffBlocks(this.diffView.getBaseline(key)?.text ?? current, current));
+      return;
+    }
+    const baseline = this.diffView.getBaseline(key);
+    const baselineText = baseline?.text ?? current;
+    const ops = diffBlocks(baselineText, current);
+    this.lastModel.set(key, ops);
+    const summary = {
+      added: ops.filter((o) => o.kind === "added").length,
+      removed: ops.filter((o) => o.kind === "removed").length,
+      changed: ops.filter((o) => o.kind === "changed").length,
+    };
+    void panel.webview.postMessage({
+      type: "render",
+      mode,
+      html: renderTrackChanges(baselineText, current),
+      epoch: this.epochLabel(baseline),
+      summary,
+    });
+  }
+```
+
+Add a test seam for the current mode (used by E2E):
+```typescript
+  getMode(uriString: string): "changes" | "authorship" {
+    return this.mode.get(uriString) ?? "changes";
+  }
+  setMode(uriString: string, mode: "changes" | "authorship"): void {
+    this.mode.set(uriString, mode);
+    const doc = vscode.workspace.textDocuments.find((d) => d.uri.toString() === uriString);
+    if (doc) this.refresh(doc);
+  }
+```
+
+- [ ] **Step 2: extension.ts — reorder + pass attribution**
+
+In `src/extension.ts`, the F7 controller is currently constructed in the F6/F7 block (before the authoring stack). Move the `TrackChangesPreviewController` construction to AFTER `attributionController` is created, and pass it:
+
+1. Delete the existing block:
+```typescript
+  const trackChangesPreviewController = new TrackChangesPreviewController(
+    diffViewController,
+    context.extensionUri,
+  );
+  context.subscriptions.push(trackChangesPreviewController);
+```
+2. Re-create it right after `const attributionController = new AttributionController(...)` + its `context.subscriptions.push(attributionController);`:
+```typescript
+  // --- F7: rendered track-changes preview (Feature #21) + F9 authorship mode ---
+  // Constructed after attribution so the authorship view can read F3 spans.
+  const trackChangesPreviewController = new TrackChangesPreviewController(
+    diffViewController,
+    context.extensionUri,
+    attributionController,
+  );
+  context.subscriptions.push(trackChangesPreviewController);
+```
+(The `CowritingApi` return already includes `trackChangesPreviewController` — unchanged. The F6 `diffViewController` stays where it is.)
+
+- [ ] **Step 3: Verify typecheck + unit**
+
+Run: `npm run typecheck && npm test`
+Expected: clean + 150+ tests green (Tasks 1–3 added tests).
+
+- [ ] **Step 4: Commit**
+
+```bash
+git add src/trackChangesPreview.ts src/extension.ts
+git commit -m "feat(f9): preview gains authorship mode + attribution dep + setMode wiring"
+```
+
+---
+
+## Task 5: Webview toggle, legend, and CSS
+
+**Files:**
+- Modify: `src/trackChangesPreview.ts` (shell HTML), `media/preview.ts`, `media/preview.css`
+
+- [ ] **Step 1: Shell HTML — add the segmented toggle + legend slot**
+
+In `src/trackChangesPreview.ts` `shellHtml(...)`, replace the `<div id="cw-header">…</div>` with:
+```html
+  <div id="cw-header">
+    <div id="cw-mode" role="group">
+      <button id="cw-mode-changes" class="cw-seg cw-seg-on" data-mode="changes">Track changes</button>
+      <button id="cw-mode-authorship" class="cw-seg" data-mode="authorship">Authorship</button>
+    </div>
+    <span id="cw-epoch">Track changes</span>
+    <span id="cw-summary"></span>
+    <span id="cw-legend" hidden></span>
+  </div>
+```
+
+- [ ] **Step 2: `media/preview.ts` — post setMode; render per mode**
+
+Replace the message interface + handler and add the toggle wiring:
+```typescript
+interface RenderMessage {
+  type: "render";
+  mode: "changes" | "authorship";
+  html: string;
+  epoch?: string;
+  summary?: { added: number; removed: number; changed: number };
+  legend?: { claude: boolean; human: boolean };
+}
+
+const vscodeApi = acquireVsCodeApi();
+const body = document.getElementById("cw-body")!;
+const header = document.getElementById("cw-epoch")!;
+const summary = document.getElementById("cw-summary")!;
+const legend = document.getElementById("cw-legend")!;
+const segs = Array.from(document.querySelectorAll<HTMLButtonElement>(".cw-seg"));
+
+for (const seg of segs) {
+  seg.addEventListener("click", () => {
+    vscodeApi.postMessage({ type: "setMode", mode: seg.dataset.mode });
+  });
+}
+
+window.addEventListener("message", (event: MessageEvent<RenderMessage>) => {
+  const msg = event.data;
+  if (msg?.type !== "render") return;
+  body.innerHTML = msg.html;
+  for (const seg of segs) seg.classList.toggle("cw-seg-on", seg.dataset.mode === msg.mode);
+  const authorship = msg.mode === "authorship";
+  header.hidden = authorship;
+  summary.hidden = authorship;
+  legend.hidden = !authorship;
+  if (authorship) {
+    const parts: string[] = [];
+    if (msg.legend?.claude) parts.push('<span class="cw-by-claude cw-swatch">Claude</span>');
+    if (msg.legend?.human) parts.push('<span class="cw-by-human cw-swatch">You</span>');
+    legend.innerHTML = parts.join(" ") || "no attribution yet";
+  } else {
+    header.textContent = `Track changes since ${msg.epoch ?? ""}`;
+    summary.innerHTML =
+      `<span class="cw-add">+${(msg.summary?.added ?? 0) + (msg.summary?.changed ?? 0)}</span> ` +
+      `<span class="cw-del">−${(msg.summary?.removed ?? 0) + (msg.summary?.changed ?? 0)}</span>`;
+  }
+  void renderMermaid();
+});
+```
+(`acquireVsCodeApi` is a webview global; add `declare function acquireVsCodeApi(): { postMessage(m: unknown): void };` near the top of the file if the type isn't already present.)
+
+- [ ] **Step 3: `media/preview.css` — author + toggle + legend styles**
+
+Append:
+```css
+.cw-by-claude { background: var(--vscode-editorInfo-foreground, rgba(64, 120, 242, 0.18)); text-decoration: none; }
+.cw-by-human { background: var(--vscode-gitDecoration-addedResourceForeground, rgba(46, 160, 67, 0.18)); text-decoration: none; }
+/* Block-level author badges reuse .cw-badge; tint the whole fence block. */
+.cw-blk.cw-by-claude, .cw-blk.cw-by-human, .cw-blk.cw-mixed { outline: 2px solid currentColor; outline-offset: 2px; }
+.cw-seg {
+  background: transparent; color: var(--vscode-foreground);
+  border: 1px solid var(--vscode-panel-border); padding: 0 0.5em; cursor: pointer; font-size: 0.9em;
+}
+.cw-seg:first-child { border-radius: 3px 0 0 3px; }
+.cw-seg:last-child { border-radius: 0 3px 3px 0; border-left: none; }
+.cw-seg-on { background: var(--vscode-button-background); color: var(--vscode-button-foreground); }
+#cw-legend .cw-swatch { padding: 0 0.4em; border-radius: 3px; }
+```
+(Background tints use translucent fallbacks so they read against any theme; the inline spans are intentionally subtle.)
+
+- [ ] **Step 4: Build + verify**
+
+Run: `npm run build && npm run typecheck && npm test`
+Expected: build emits `out/media/preview.js`; typecheck clean; unit green.
+
+- [ ] **Step 5: Commit**
+
+```bash
+git add src/trackChangesPreview.ts media/preview.ts media/preview.css
+git commit -m "feat(f9): webview segmented mode toggle + authorship legend + author CSS"
+```
+
+---
+
+## Task 6: Host E2E
+
+**Files:**
+- Create: `test/e2e/suite/authorship.test.ts`
+
+- [ ] **Step 1: Write the suite**
+
+```typescript
+import * as assert from "assert";
+import * as fs from "fs";
+import * as path from "path";
+import * as vscode from "vscode";
+import type { CowritingApi } from "../../../src/extension";
+
+const WS = process.env.E2E_WORKSPACE!;
+const settle = () => new Promise((r) => setTimeout(r, 300));
+
+async function getApi(): Promise<CowritingApi> {
+  const ext = vscode.extensions.getExtension("benstull.vscode-cowriting-plugin")!;
+  const api = (await ext.activate()) as CowritingApi;
+  assert.ok(api?.trackChangesPreviewController && api?.proposalController, "exports preview + proposal");
+  return api;
+}
+
+// F9 host E2E (no LLM): authorship mode marks Claude's landed span. Owns its own
+// markdown doc, disjoint from the other suites' fixtures.
+suite("F9 authorship preview (host E2E — seam ingress, no LLM)", () => {
+  const DOC_REL = "docs/f9authorship.md";
+  const TARGET = "The sentence Claude will compose over.";
+
+  test("authorship mode marks Claude's accepted edit; track-changes mode still works", async () => {
+    const abs = path.join(WS, DOC_REL);
+    fs.mkdirSync(path.dirname(abs), { recursive: true });
+    fs.writeFileSync(abs, `# F9\n\n${TARGET}\n`, "utf8");
+    const uri = vscode.Uri.file(abs);
+    const doc = await vscode.workspace.openTextDocument(uri);
+    await vscode.window.showTextDocument(doc);
+    await settle();
+    const api = await getApi();
+    const key = uri.toString();
+
+    // open the preview (track-changes mode by default)
+    await vscode.commands.executeCommand("cowriting.showTrackChangesPreview");
+    await settle();
+    assert.ok(api.trackChangesPreviewController.isOpen(key), "preview open");
+    assert.strictEqual(api.trackChangesPreviewController.getMode(key), "changes", "defaults to track-changes");
+
+    // Claude composes via the seam (propose → accept)
+    const start = doc.getText().indexOf(TARGET);
+    const id = await vscode.commands.executeCommand<string>("cowriting.proposeAgentEdit", {
+      uri: key, start, end: start + TARGET.length, newText: "The sentence CLAUDE COMPOSED.", model: "sonnet", sessionId: "e2e-f9", turnId: "turn-f9",
+    });
+    assert.ok(await api.proposalController.acceptById(DOC_REL, id!), "accept applies");
+    await settle();
+
+    // attribution has a Claude span now
+    const claudeSpan = api.attributionController.getSpans(DOC_REL).find((s) => s.authorKind === "agent");
+    assert.ok(claudeSpan, "Claude span recorded by F3");
+
+    // flip to authorship mode → the model should reflect Claude authorship
+    api.trackChangesPreviewController.setMode(key, "authorship");
+    await settle();
+    assert.strictEqual(api.trackChangesPreviewController.getMode(key), "authorship");
+    const spans = api.attributionController.spansFor(doc);
+    assert.ok(spans.some((s) => s.author === "claude"), "spansFor reports a Claude span for the preview");
+
+    // back to track-changes — still functional (regression)
+    api.trackChangesPreviewController.setMode(key, "changes");
+    await settle();
+    assert.strictEqual(api.trackChangesPreviewController.getMode(key), "changes");
+    assert.ok((api.trackChangesPreviewController.getLastModel(key) ?? []).length >= 1, "track-changes model still computed");
+  });
+});
+```
+
+- [ ] **Step 2: Run E2E**
+
+Run: `npm run test:e2e`
+Expected: both EDH passes green (the new F9 suite + all prior suites). Debug per `superpowers:systematic-debugging` if red; do not weaken assertions.
+
+- [ ] **Step 3: Commit**
+
+```bash
+git add test/e2e/suite/authorship.test.ts
+git commit -m "test(f9): host E2E — authorship mode marks Claude's landed span"
+```
+
+---
+
+## Task 7: Docs
+
+**Files:**
+- Create: `docs/MANUAL-SMOKE-F9.md`
+- Modify: `README.md`
+
+- [ ] **Step 1: `docs/MANUAL-SMOKE-F9.md`**
+
+```markdown
+# Manual smoke — F9 authorship view in the preview
+
+Confirms the rendered preview's Authorship mode colors Claude's vs your text. One
+live turn hits the SDK.
+
+## Prereqs
+- `npm run build`; launch the Extension Development Host (F5) with `sandbox/` open.
+
+## Steps
+1. Open a markdown file. Open **Cowriting: Open Track-Changes Preview** (`Ctrl+Alt+R`).
+2. The header shows a `[ Track changes | Authorship ]` toggle. It opens in **Track
+   changes** (unchanged behavior).
+3. Select a sentence → **Ask Claude to Edit Selection** → instruct → accept.
+4. Click **Authorship**. Expect: Claude's accepted text is tinted (blue) and your
+   own typing is tinted (green); the header shows a legend (● Claude / ● You).
+   Text you never touched (original content) is plain.
+5. Type a few words yourself, mixing into a Claude sentence. Expect: the colors
+   split mid-paragraph at the exact boundaries.
+6. If the doc has a mermaid or code fence Claude authored, expect the whole fence
+   to carry a small author badge (not per-character).
+7. Flip back to **Track changes**. Expect: the diff view is exactly as before.
+```
+
+- [ ] **Step 2: README F9 note** — add after the F8 section (match heading style):
+
+```markdown
+## F9 — Authorship view in the preview (Feature ~#27)
+
+The rendered preview (F7) gains a second mode, switched by a `[ Track changes |
+Authorship ]` toggle in its header. **Authorship** mode re-renders the current
+document with each span colored by its F3 author — Claude (blue) vs you (green),
+inline and char-precise — with a legend. Unlike track-changes (which diffs against
+the F6 baseline, and so hides Claude's text once the baseline advances past a
+landing), authorship reads F3 attribution directly, so Claude's contributions stay
+visible. Code/mermaid fences carry a block-level author badge (atomic). Read-only,
+sealed webview, no new persistence (INV-26..28).
+
+Design: `docs/superpowers/specs/2026-06-11-authorship-preview-design.md`.
+Live smoke: [`docs/MANUAL-SMOKE-F9.md`](docs/MANUAL-SMOKE-F9.md).
+```
+
+- [ ] **Step 3: Verify + commit**
+
+```bash
+npm run typecheck && npm test
+git add docs/MANUAL-SMOKE-F9.md README.md
+git commit -m "docs(f9): manual smoke + README authorship-mode note"
+```
+
+---
+
+## Self-Review checklist (run before PR)
+
+- **Spec coverage:** §2 modes → Tasks 4/5; §3.1 render engine → Tasks 1/2; §3.2 wiring → Tasks 3/4/5; §3.3 INV-26/27/28 → Tasks 2/4; §5 testing → Tasks 1/2/6 + smoke; §6 slices → Tasks 1–7.
+- **Untouched:** `renderTrackChanges`/`diffBlocks` logic, F6 baseline, F3 capture, the seam, persistence (`git diff --stat` should show no behavioral change to those).
+- **Done (spec §6):** authorship mode marks Claude (blue) + you (green) inline; header toggle flips modes; fences get a block badge; track-changes unchanged; unit + host E2E green; smoke performed once.
+```
@@ -0,0 +1,123 @@
+# Implementation Plan: F11 — Preview Toolbar as the Primary Interaction Surface (#43)
+
+**Spec:** `docs/superpowers/specs/2026-06-12-f11-preview-toolbar-interaction-surface.md`
+**Anchor:** Feature `benstull/vscode-cowriting-plugin#43` (F11, `type/feature`, `priority/P1`)
+**Session:** vscode-cowriting-plugin-0037
+
+This plan transcribes the spec's §7.2 slicing plan into concrete, file-level
+tasks. Each slice is independently green (unit + host E2E) before the next. Host
+E2E is this app's required tier (no browser/deploy stage — a VS Code extension);
+no LLM in CI (edit turns stubbed). The webview's visual rendering, the adaptive
+label, and the selection→source DOM lookup are manual-smoke only.
+
+---
+
+## SLICE-1 — Pin baseline button + reachability *(the immediate win)*
+
+Homes the orphaned `cowriting.pinDiffBaseline` command and gives the writer a
+reachable Pin control in the preview toolbar.
+
+**Tasks**
+
+1. **Host message routing** (`src/trackChangesPreview.ts`): extract the inline
+   `onDidReceiveMessage` body into a private `handleWebviewMessage(document, m)`
+   method; add a `pinBaseline` branch that calls `this.diffView.pin(document)`
+   (the *previewed* document — not `activeTextEditor`). The existing
+   `onDidChangeBaseline` subscription already re-renders with cleared marks.
+2. **Test seam**: add `receiveMessage(uriString, m)` that resolves the doc and
+   calls `handleWebviewMessage`, so host E2E can simulate the raw webview message
+   and exercise the real routing.
+3. **Webview** (`media/preview.ts` + `.css`): add a `⌖ Pin baseline` button in
+   `#cw-header`; click → `postMessage({ type: "pinBaseline" })`; theme-aware CSS.
+4. **Shell HTML** (`shellHtml`): add the `<button id="cw-pin">` to the header row.
+5. **Reachability** (`package.json`): flip `cowriting.pinDiffBaseline`'s
+   `commandPalette` `when` from `false` to `editorLangId == markdown`.
+
+**Host E2E** (`test/e2e/suite/trackChangesPreview.test.ts` or new f11 suite):
+open a markdown fixture with a divergent baseline → some block marked; simulate
+`{type:"pinBaseline"}` via `receiveMessage` → `getLastModel` shows every block
+`unchanged` (marks cleared) and `getBaseline(key).reason === "pinned"`.
+
+**Unit:** none new (pure layer untouched this slice).
+
+---
+
+## SLICE-2 — Block-offset emission *(INV-36 data layer)*
+
+Shared pure helper wrapping each rendered block with `data-src-start`/`-end`
+(source char offsets from `BlockWithRange`) in **both** `renderReview` and
+`renderPlain`. vscode-free, DOM-free, deterministic (extends INV-22/33). No UI.
+
+**Tasks**
+
+1. `src/trackChangesModel.ts`: a shared internal helper that prepends
+   `data-src-start="N" data-src-end="M"` to each block's wrapping element, routed
+   from both render paths using the existing `splitBlocksWithRanges` offsets.
+2. Skip blocks with no live-source range (deletion-only / proposal blocks).
+
+**Unit** (`test/trackChangesModel.test.ts`): `data-src-start/end` present and
+correct on every block for both modes (offsets equal `BlockWithRange` ranges);
+determinism (same inputs → identical HTML).
+
+---
+
+## SLICE-3 — Edit Document button + hunk path *(INV-37 document half)*
+
+**Tasks**
+
+1. `src/trackChangesModel.ts`: pure `diffToHunks(currentText, rewrittenText):
+   Array<{ start; end; replacement }>` — vscode-free, deterministic.
+2. `src/trackChangesPreview.ts`: `runEditAndPropose(document, target, instruction)`
+   private routine; `askClaude`/`document` branch → host `showInputBox` →
+   `runEditTurn` over full text → `diffToHunks` → one F4 `propose()` per hunk.
+3. `package.json`: register `cowriting.editDocument` (document-scoped), routed
+   through `runEditAndPropose({kind:"document"})`; for `#42` reuse.
+4. Webview: `✦ Ask Claude to Edit Document` button (no-selection state) →
+   `postMessage({ type: "askClaude", scope: "document" })`.
+
+**Unit:** `diffToHunks` over fixtures (single hunk → one range; multi-hunk →
+disjoint ranges + correct replacements; unchanged → zero hunks; whole-doc
+replacement → one full-range hunk).
+
+**Host E2E:** simulate `{askClaude, scope:"document"}` with a stubbed multi-hunk
+rewrite → **N** proposals matching the hunks.
+
+---
+
+## SLICE-4 — Adaptive Edit Selection *(INV-37 selection half; INV-36 consumer)*
+
+**Tasks**
+
+1. Webview: `selectionchange` listener flips the Ask-Claude label (Edit Selection
+   ⇆ Edit Document); selection→nearest-`data-src` ancestor resolution →
+   `postMessage({ type:"askClaude", scope:"selection", start, end })`.
+2. Host: `askClaude`/`selection` branch → `runEditAndPropose({kind:"range",
+   start, end})` → one `runEditTurn` → one F4 `propose()` over the block-union.
+3. Edge: selection resolving to no live block → fall back to document scope.
+
+**Host E2E:** simulate `{askClaude, scope:"selection", start, end}` with a stubbed
+edit turn → exactly **one** proposal over the resolved range, anchored inline.
+
+---
+
+## SLICE-5 — Gateway, edges, tests & docs
+
+**Tasks**
+
+1. `package.json`: add `cowriting.showTrackChangesPreview` to `editor/title` with
+   `when: editorLangId == markdown` (minimal right-click gateway).
+2. Non-authorable disabling: Pin + Ask-Claude controls render disabled when
+   `!isAuthorable(document)`; annotations toggle stays active.
+3. Host E2E: gateway command opens the panel; controls inert on non-authorable.
+4. `docs/MANUAL-SMOKE-F11.md` (live smoke script per spec §6.8).
+5. README F11 section.
+
+---
+
+## Done = #43 acceptance (spec §7.3)
+
+Preview toolbar hosts annotations checkbox + Pin baseline + single adaptive
+Ask-Claude (Edit Selection ⇆ Edit Document) routing through existing F4/F3/F6;
+edits surface as proposals (one for a selection, per-hunk for a document
+rewrite); a right-click entry opens the preview; the pin command is no longer
+orphaned; unit + host E2E green; live smoke performed once.
@@ -0,0 +1,827 @@
+# Author-colored track changes Implementation Plan
+
+> **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.
+
+**Goal:** Render every change as a track-changes diff where **style = operation** (underline = inserted, strikethrough = removed) and **color = author** (human green/red, Claude blue/purple), in both the rendered preview pane and the main editor pane.
+
+**Architecture:** Fuse the two color systems that exist today — the *semantic* diff (green=add/red=remove) and the *authorship* tint (cw-by-claude/cw-by-human) — into one author-colored diff. Insertions are colored by their attribution span (data already exists). Deletions are colored by an **adjacency heuristic**: a struck run inherits the author of the insertion it is paired with in the same changed hunk; a standalone removed block falls back to neutral. Phase 1 does this in the pure render engine (preview). Phase 2 reverses INV-32 to bring the same author-colored track changes inline into the editor.
+
+**Tech Stack:** TypeScript, VS Code extension API, `markdown-it`, `diff` (jsdiff), vitest (unit), Playwright/electron host harness (E2E). Render engine (`src/trackChangesModel.ts`) is pure and vscode-free.
+
+## Global Constraints
+
+- **Render engine stays pure / vscode-free / deterministic** — `src/trackChangesModel.ts` must not import `vscode`; same inputs → identical HTML (INV-22).
+- **Author color mapping (verbatim):** human added `#3fb950` / human removed `#f85149` / Claude added `#58a6ff` / Claude removed `#bc8cff`. Insert = 2px underline in the author color + faint author-hued background; remove = line-through in the author color + faint background.
+- **Convention:** underline = inserted, strikethrough = removed, color = author.
+- **"Color means changed since the pinned baseline."** Unchanged-since-baseline text renders plain (no author coloring). The pin→clean behavior (#48) is preserved: a pinned, zero-diff doc renders with no annotation.
+- **Deletion authorship = adjacency heuristic** (locked decision): paired deletions inherit the paired insertion's author; standalone deletions render neutral.
+- **Two roles only:** `human` and `claude` (`AuthorKind`). No arbitrary palettes, no user-configurable colors (YAGNI).
+- **Test command:** `npm test` (vitest, unit). E2E: `npm run test:e2e`. Typecheck: `npm run typecheck`.
+- Commit after each task. Branch off `main` (do not commit to `main` directly).
+
+---
+
+# Phase 1 — Preview pane (independently shippable)
+
+Delivers the full author-colored track-changes language in the rendered preview, where the attribution data already exists.
+
+## File Structure (Phase 1)
+
+- `src/trackChangesModel.ts` — add `authorAt()` + `wordDiffByAuthor()` pure helpers; route the review path through them; color proposal blocks by author; stop coloring unchanged blocks. Add `author?` to `ProposalView`.
+- `src/trackChangesPreview.ts` — pass each proposal's author into its `ProposalView`.
+- `media/preview.css` — author×operation classes; retire the semantic green/red and flat authorship tints.
+- `test/trackChangesModel.test.ts` — new unit tests; update tests asserting old semantic classes.
+
+---
+
+### Task 1: `authorAt()` + `wordDiffByAuthor()` pure helpers
+
+**Files:**
+- Modify: `src/trackChangesModel.ts` (add after `wordMergedMarkdown`, ~line 475)
+- Test: `test/trackChangesModel.test.ts`
+
+**Interfaces:**
+- Consumes: existing `AuthorKind`, `AuthorSpan` (trackChangesModel.ts:557-562), `diffWordsWithSpace` (already imported, line 12).
+- Produces:
+  - `authorAt(offset: number, spans: AuthorSpan[]): AuthorKind | null` — the author whose span covers `offset` (half-open), else null.
+  - `wordDiffByAuthor(beforeRaw: string, afterRaw: string, afterStart: number, spans: AuthorSpan[]): string` — inline HTML where each inserted run is `<ins class="cw-ins-{author}">` (author looked up at the run's `afterStart`-relative landed offset) and each removed run is `<del class="cw-del-{author}">` (author = the insertion it's paired with in the same change cluster; `cw-del-none` if standalone). Unchanged parts are emitted verbatim.
+
+- [ ] **Step 1: Write the failing test**
+
+Add to `test/trackChangesModel.test.ts` (import `authorAt`, `wordDiffByAuthor` from `../src/trackChangesModel`):
+
+```ts
+describe("authorAt", () => {
+  const spans: AuthorSpan[] = [
+    { start: 0, end: 5, author: "human" },
+    { start: 5, end: 10, author: "claude" },
+  ];
+  test("returns the covering span's author (half-open)", () => {
+    expect(authorAt(0, spans)).toBe("human");
+    expect(authorAt(4, spans)).toBe("human");
+    expect(authorAt(5, spans)).toBe("claude");
+    expect(authorAt(9, spans)).toBe("claude");
+  });
+  test("returns null outside any span", () => {
+    expect(authorAt(10, spans)).toBeNull();
+    expect(authorAt(-1, spans)).toBeNull();
+  });
+});
+
+describe("wordDiffByAuthor", () => {
+  test("an inserted run is colored by the author covering its landed offset", () => {
+    // before "hello " ; after "hello brave " ; "brave " inserted at landed offset 6..12
+    const spans: AuthorSpan[] = [{ start: 6, end: 12, author: "claude" }];
+    const html = wordDiffByAuthor("hello ", "hello brave ", 0, spans);
+    expect(html).toContain('<ins class="cw-ins-claude">brave </ins>');
+    expect(html).toContain("hello ");
+  });
+  test("a paired deletion inherits the paired insertion's author (adjacency)", () => {
+    // "light" -> "dark", both in one cluster; "dark" is claude-inserted
+    const spans: AuthorSpan[] = [{ start: 0, end: 4, author: "claude" }];
+    const html = wordDiffByAuthor("light", "dark", 0, spans);
+    expect(html).toContain('<del class="cw-del-claude">light</del>');
+    expect(html).toContain('<ins class="cw-ins-claude">dark</ins>');
+  });
+  test("a standalone deletion (no paired insertion) is neutral", () => {
+    const html = wordDiffByAuthor("keep this word", "keep word", 0, []);
+    expect(html).toContain('<del class="cw-del-none">this </del>');
+    expect(html).not.toContain("cw-del-human");
+    expect(html).not.toContain("cw-del-claude");
+  });
+});
+```
+
+- [ ] **Step 2: Run test to verify it fails**
+
+Run: `npm test -- trackChangesModel`
+Expected: FAIL — `authorAt`/`wordDiffByAuthor` not exported.
+
+- [ ] **Step 3: Write minimal implementation**
+
+Add to `src/trackChangesModel.ts` after `wordMergedMarkdown` (line 475):
+
+```ts
+/** The author whose span covers `offset` (half-open [start,end)), or null. */
+export function authorAt(offset: number, spans: AuthorSpan[]): AuthorKind | null {
+  for (const s of spans) if (offset >= s.start && offset < s.end) return s.author;
+  return null;
+}
+
+const insClass = (a: AuthorKind | null): string => `cw-ins-${a ?? "none"}`;
+const delClass = (a: AuthorKind | null): string => `cw-del-${a ?? "none"}`;
+
+/**
+ * Author-colored word diff for a changed PROSE block in the review.
+ * `afterStart` is the landed-text offset of `afterRaw[0]` so an inserted run's
+ * offset maps into `spans`. Insertions are colored by the author covering their
+ * offset; deletions inherit the author of the insertion they are paired with in
+ * the same contiguous change cluster (adjacency heuristic) — `cw-del-none` when a
+ * cluster has no insertion. Pure, deterministic.
+ */
+export function wordDiffByAuthor(
+  beforeRaw: string,
+  afterRaw: string,
+  afterStart: number,
+  spans: AuthorSpan[],
+): string {
+  const parts = diffWordsWithSpace(beforeRaw, afterRaw);
+  // First pass: find each change cluster's insertion author (first added run).
+  // A cluster is a maximal run of added/removed parts bounded by unchanged parts.
+  let afterOff = afterStart;
+  const clusterAuthor: (AuthorKind | null)[] = []; // per part index, the cluster's del author
+  {
+    let off = afterStart;
+    let i = 0;
+    while (i < parts.length) {
+      const p = parts[i];
+      if (!p.added && !p.removed) {
+        clusterAuthor[i] = null;
+        off += p.value.length;
+        i++;
+        continue;
+      }
+      // a cluster: scan to its end, capturing the first added run's author
+      let j = i;
+      let author: AuthorKind | null = null;
+      let scanOff = off;
+      while (j < parts.length && (parts[j].added || parts[j].removed)) {
+        if (parts[j].added && author === null) author = authorAt(scanOff, spans);
+        if (parts[j].added) scanOff += parts[j].value.length;
+        j++;
+      }
+      for (let k = i; k < j; k++) clusterAuthor[k] = author;
+      off = scanOff;
+      i = j;
+    }
+  }
+  // Second pass: emit.
+  const out: string[] = [];
+  parts.forEach((p, i) => {
+    if (p.added) {
+      const a = authorAt(afterOff, spans);
+      out.push(`<ins class="${insClass(a)}">${p.value}</ins>`);
+      afterOff += p.value.length;
+    } else if (p.removed) {
+      out.push(`<del class="${delClass(clusterAuthor[i] ?? null)}">${p.value}</del>`);
+    } else {
+      out.push(p.value);
+      afterOff += p.value.length;
+    }
+  });
+  return out.join("");
+}
+```
+
+- [ ] **Step 4: Run test to verify it passes**
+
+Run: `npm test -- trackChangesModel`
+Expected: PASS (new describes green).
+
+- [ ] **Step 5: Commit**
+
+```bash
+git add src/trackChangesModel.ts test/trackChangesModel.test.ts
+git commit -m "feat: authorAt + wordDiffByAuthor pure helpers (author-colored word diff w/ adjacency del)"
+```
+
+---
+
+### Task 2: Route the review path through `wordDiffByAuthor`; stop coloring unchanged blocks
+
+**Files:**
+- Modify: `src/trackChangesModel.ts` — `renderReviewOp` (769-780), the `colored` closure + changed-op handling in `renderReview` (889-897)
+- Test: `test/trackChangesModel.test.ts`
+
+**Interfaces:**
+- Consumes: `wordDiffByAuthor` (Task 1), `landedSpans` + `ranges` already computed in `renderReview`.
+- Produces: review HTML where changed prose blocks emit `cw-ins-{author}`/`cw-del-{author}`, added blocks are author-colored as insertions, and **unchanged blocks render plain**.
+
+- [ ] **Step 1: Write the failing test**
+
+Add to the `renderReview` describe in `test/trackChangesModel.test.ts`:
+
+```ts
+test("renderReview: a changed prose block colors ins/del by author (claude replace)", () => {
+  // baseline "light" -> current "dark"; "dark" attributed to claude at 0..4
+  const html = renderReview("light", "dark", [{ start: 0, end: 4, author: "claude" }], []);
+  expect(html).toContain('cw-ins-claude');
+  expect(html).toContain('cw-del-claude'); // adjacency: struck "light" inherits claude
+});
+
+test("renderReview: an unchanged block renders plain (no author coloring)", () => {
+  const doc = "Alpha stays.";
+  const html = renderReview(doc, doc, [{ start: 0, end: 12, author: "human" }], []);
+  expect(html).not.toContain("cw-by-");
+  expect(html).not.toContain("cw-ins-");
+});
+
+test("renderReview: an added block is author-colored as an insertion", () => {
+  // baseline empty-ish -> add a new paragraph authored by human
+  const html = renderReview("Keep.", "Keep.\n\nFresh line.", [{ start: 6, end: 17, author: "human" }], []);
+  expect(html).toContain("cw-ins-human");
+});
+```
+
+- [ ] **Step 2: Run test to verify it fails**
+
+Run: `npm test -- trackChangesModel`
+Expected: FAIL — changed blocks still emit plain `<ins>/<del>` via `wordMergedMarkdown`; unchanged blocks still carry `cw-by-*`.
+
+- [ ] **Step 3: Write minimal implementation**
+
+In `src/trackChangesModel.ts`:
+
+**(a)** Change `renderReviewOp` (769-780) so a changed PROSE block uses author coloring. Give it the colored-insertion path for `added` and the author word-diff for non-atomic `changed`. Replace the function body:
+
+```ts
+function renderReviewOp(
+  op: BlockOp,
+  render: (src: string) => string,
+  colored: (raw: string) => string,
+  src: string,
+  changedHtml?: string,
+): string {
+  // A non-atomic changed prose block is pre-rendered with author-colored ins/del
+  // (changedHtml). Removed blocks and atomic changed fences stay neutral via renderOp
+  // (standalone deletion → neutral, per the adjacency heuristic). `src` is "" for a
+  // removed block (no live source — INV-36).
+  if (op.kind === "changed" && !op.atomic && changedHtml !== undefined) {
+    return `<div class="cw-blk cw-changed"${src}>${changedHtml}</div>`;
+  }
+  if (op.kind === "removed" || op.kind === "changed") return renderOp(op, render, src);
+  return `<div class="cw-blk ${op.kind === "added" ? "cw-added" : "cw-unchanged"}"${src}>${colored(op.block.raw)}</div>`;
+}
+```
+
+**(b)** In `renderReview` (the loop at 889-897), compute the author-colored changed HTML and stop coloring unchanged blocks. Replace the loop body:
+
+```ts
+  let ci = 0; // pointer into ranges; advances for every op with a current-side block
+  const bodyParts: string[] = [];
+  for (const op of ops) {
+    const blockIndex = op.kind === "removed" ? -1 : ci;
+    const blk = op.kind === "removed" ? undefined : ranges[ci++];
+    // Only ADDED blocks are author-colored (they are insertions since baseline);
+    // UNCHANGED blocks render plain (color means "changed since baseline").
+    const colored = (raw: string): string =>
+      blk && !clean && op.kind === "added" ? colorByAuthor(raw, blk.start, landedSpans, render) : render(raw);
+    // A non-atomic changed prose block: author-colored word diff (ins by author,
+    // del by adjacency). `blk.start` is the landed offset of the after-side text.
+    const changedHtml =
+      op.kind === "changed" && !op.atomic && blk && !clean
+        ? render(wordDiffByAuthor(op.before.raw, op.block.raw, blk.start, landedSpans))
+        : undefined;
+    bodyParts.push(renderReviewOp(op, render, colored, srcAttr(blk), changedHtml));
+    const here = blockIndex >= 0 ? byBlock.get(blockIndex) : undefined;
+    if (here) for (const p of here) bodyParts.push(proposalBlockHtml(p, render));
+  }
+  for (const p of trailing) bodyParts.push(proposalBlockHtml(p, render));
+  return bodyParts.join("\n");
+```
+
+> Note: `render(wordDiffByAuthor(...))` passes the HTML-bearing string through markdown-it (same as `wordMergedMarkdown` was rendered). `colorByAuthor` on added blocks emits `cw-by-{author}` spans — Task 4 restyles those to insert styling in CSS.
+
+- [ ] **Step 4: Run test to verify it passes**
+
+Run: `npm test -- trackChangesModel`
+Expected: PASS for the new tests.
+
+- [ ] **Step 5: Update tests that assert the OLD review markup**
+
+Two existing tests assert the old semantic output (test/trackChangesModel.test.ts:226-232). Update them:
+
+```ts
+test("renderReview: human addition since baseline renders author-colored ins", () => {
+  const html = renderReview("hello", "hello world", [{ start: 6, end: 11, author: "human" }], []);
+  expect(html).toContain("cw-ins-human");
+});
+test("renderReview: a standalone deletion since baseline renders neutral struck del", () => {
+  const html = renderReview("hello world", "hello", [], []);
+  expect(html).toMatch(/cw-del-none|cw-removed/);
+});
+```
+
+Run: `npm test -- trackChangesModel`
+Expected: PASS (whole file green).
+
+- [ ] **Step 6: Commit**
+
+```bash
+git add src/trackChangesModel.ts test/trackChangesModel.test.ts
+git commit -m "feat: author-colored changed/added blocks in review; unchanged blocks render plain"
+```
+
+---
+
+### Task 3: Color pending proposal blocks by author
+
+**Files:**
+- Modify: `src/trackChangesModel.ts` — `ProposalView` (731-742) + `proposalBlockHtml` (744-767)
+- Modify: `src/trackChangesPreview.ts` — populate `author` on each `ProposalView`
+- Test: `test/trackChangesModel.test.ts`
+
+**Interfaces:**
+- Consumes: `AuthorKind` (557).
+- Produces: `ProposalView.author?: AuthorKind` (defaults to `"claude"`); `proposalBlockHtml` emits `cw-del-{author}` / `cw-ins-{author}`.
+
+- [ ] **Step 1: Write the failing test**
+
+```ts
+test("renderReview: a proposal block colors its del/ins by author (claude default)", () => {
+  const proposals: ProposalView[] = [{ id: "p1", anchorStart: 0, anchorEnd: 5, replaced: "hello", replacement: "goodbye" }];
+  const html = renderReview("hello", "hello", [], proposals);
+  expect(html).toContain('cw-del-claude');
+  expect(html).toContain('cw-ins-claude');
+});
+```
+
+- [ ] **Step 2: Run test to verify it fails**
+
+Run: `npm test -- trackChangesModel`
+Expected: FAIL — `proposalBlockHtml` still emits `cw-del` / `cw-add`.
+
+- [ ] **Step 3: Write minimal implementation**
+
+In `src/trackChangesModel.ts`, add `author?` to `ProposalView` (after line 741):
+
+```ts
+  /** F12/#64 (INV-48): the pre-apply original, set after optimistic apply. */
+  original?: string;
+  /** Who made the proposal — colors the del/ins. Defaults to "claude". */
+  author?: AuthorKind;
+}
+```
+
+Update `proposalBlockHtml` (753-754):
+
+```ts
+  const who = p.author ?? "claude";
+  const before = p.replaced ? `<del class="cw-del-${who}">${safe(p.replaced)}</del>` : "";
+  const after = `<ins class="cw-ins-${who}">${safe(p.replacement)}</ins>`;
+```
+
+- [ ] **Step 4: Run test to verify it passes**
+
+Run: `npm test -- trackChangesModel`
+Expected: PASS.
+
+- [ ] **Step 5: Populate `author` from the live proposal**
+
+In `src/trackChangesPreview.ts` find where `ProposalView`s are built for `renderReview` (the `listProposals` mapping near line 373). Add `author` from the proposal's provenance. Locate the mapping and add:
+
+```ts
+      author: p.author?.kind === "agent" ? "claude" : "human",
+```
+
+(where `p` is the live `Proposal` with `author: Provenance`). If the preview builds `ProposalView` via a helper, add the field there. Run `npm run typecheck` to confirm the field name matches the live `Proposal.author` shape (`model.ts:79-108`).
+
+- [ ] **Step 6: Run typecheck + tests**
+
+Run: `npm run typecheck && npm test -- trackChangesModel`
+Expected: PASS.
+
+- [ ] **Step 7: Commit**
+
+```bash
+git add src/trackChangesModel.ts src/trackChangesPreview.ts test/trackChangesModel.test.ts
+git commit -m "feat: color pending proposal blocks by author (Claude blue/purple)"
+```
+
+---
+
+### Task 4: CSS — author×operation color system (preview)
+
+**Files:**
+- Modify: `media/preview.css`
+
+**Interfaces:**
+- Consumes: classes emitted by Tasks 1-3: `cw-ins-human|claude|none`, `cw-del-human|claude|none`, `cw-by-human|claude` (added blocks), `cw-added`, `cw-removed`, `cw-changed`, `.cw-proposal`.
+- Produces: the locked color mapping; unchanged text plain.
+
+- [ ] **Step 1: Replace the semantic diff + authorship-tint rules**
+
+In `media/preview.css`, replace lines 21-33 and 55-58 with author-colored rules. New block:
+
+```css
+/* Author-colored track changes — style = operation, color = author.
+   underline = inserted, strikethrough = removed; human green/red, Claude blue/purple. */
+#cw-summary .cw-add { color: #3fb950; }
+#cw-summary .cw-del { color: #f85149; }
+.cw-blk { position: relative; }
+
+/* base ins/del carry the operation; author classes carry the color */
+ins { text-decoration: none; }
+del { text-decoration: line-through; opacity: 0.7; }
+
+.cw-ins-human  { background: rgba(63,185,80,0.14);  border-bottom: 2px solid #3fb950; text-decoration: none; }
+.cw-ins-claude { background: rgba(88,166,255,0.15); border-bottom: 2px solid #58a6ff; text-decoration: none; }
+.cw-ins-none   { background: var(--vscode-diffEditor-insertedTextBackground, rgba(63,185,80,0.14)); text-decoration: none; }
+.cw-del-human  { background: rgba(248,81,73,0.11);  text-decoration: line-through; text-decoration-color: #f85149; }
+.cw-del-claude { background: rgba(188,140,255,0.13); text-decoration: line-through; text-decoration-color: #bc8cff; }
+.cw-del-none   { background: var(--vscode-diffEditor-removedTextBackground, rgba(248,81,73,0.11)); text-decoration: line-through; opacity: 0.7; }
+
+/* whole-block ops: an added block is an insertion (author-colored inside via cw-by-*);
+   a standalone removed block is neutral struck (adjacency heuristic fallback) */
+.cw-added   { background: transparent; }
+.cw-removed { background: var(--vscode-diffEditor-removedTextBackground, rgba(248,81,73,0.10)); text-decoration: line-through; opacity: 0.65; }
+.cw-changed { outline: none; }
+
+/* added-block author runs (colorByAuthor sentinels) read as insertions */
+.cw-by-human  { background: rgba(63,185,80,0.14);  border-bottom: 2px solid #3fb950; text-decoration: none; }
+.cw-by-claude { background: rgba(88,166,255,0.15); border-bottom: 2px solid #58a6ff; text-decoration: none; }
+.cw-blk.cw-by-claude, .cw-blk.cw-by-human, .cw-blk.cw-mixed { outline: 2px solid currentColor; outline-offset: 2px; background: transparent; }
+#cw-legend .cw-swatch { padding: 0 0.4em; border-radius: 3px; }
+#cw-summary .cw-prop { opacity: 0.85; }
+```
+
+- [ ] **Step 2: Recolor the proposal block accent to neutral (author lives inside it now)**
+
+The proposal block's left border is currently blue (`--vscode-charts-blue`). Keep it as a neutral "pending" frame — the del/ins inside now carry the author color. Change `media/preview.css:79-80`:
+
+```css
+  border-left: 3px solid var(--vscode-panel-border, #555);
+  background: color-mix(in srgb, var(--vscode-foreground) 5%, transparent);
+```
+
+- [ ] **Step 3: Manual visual verification**
+
+Run the extension host (or load the smoke doc) and confirm in the preview: human insertions green-underlined, Claude insertions blue-underlined, Claude deletions purple-struck, standalone deletions neutral, unchanged text plain.
+
+Run: `npm run build`
+Expected: builds clean (CSS is bundled into the webview assets per the build).
+
+- [ ] **Step 4: Commit**
+
+```bash
+git add media/preview.css
+git commit -m "feat: author-colored track-changes CSS for the preview (human green/red, Claude blue/purple)"
+```
+
+---
+
+### Task 5: Phase-1 host E2E — author colors render in the preview
+
+**Files:**
+- Test: the existing F10/F7 preview E2E suite (find under `test/e2e/` — the preview/review specs)
+
+**Interfaces:**
+- Consumes: the rendered preview webview HTML.
+
+- [ ] **Step 1: Write the failing E2E assertions**
+
+In the preview E2E spec, after producing a Claude change + a human change on a doc, assert the webview body contains `cw-ins-claude` and `cw-ins-human`, and that an unchanged paragraph has no `cw-ins-`/`cw-by-` class. (Follow the existing pattern in that spec for opening the preview and reading `panel.webview` HTML or the DOM via the harness.)
+
+- [ ] **Step 2: Run to verify it fails (or update an assertion that checked old classes)**
+
+Run: `npm run test:e2e`
+Expected: FAIL if any existing assertion checked `cw-add`/`cw-by-*`-as-authorship; update those to the new classes.
+
+- [ ] **Step 3: Make it pass**
+
+Adjust any E2E assertion referencing retired classes (`cw-add`, semantic `ins`/`del` colors, `cw-by-*` on unchanged blocks) to the new author classes.
+
+Run: `npm run test:e2e`
+Expected: PASS.
+
+- [ ] **Step 4: Commit**
+
+```bash
+git add test/e2e
+git commit -m "test(e2e): assert author-colored track changes render in the preview"
+```
+
+**>>> Phase 1 is independently shippable here. Open a PR for the preview pane before starting Phase 2, or continue.**
+
+---
+
+# Phase 2 — Editor pane (reverses INV-32)
+
+Brings the same author-colored track changes inline into the main editor: all changes-since-baseline (committed + pending), not just pending proposals.
+
+## File Structure (Phase 2)
+
+- `src/editorProposalController.ts` — per-author decoration types; recolor proposals; subscribe to baseline + attribution; decorate all committed insertions (author-colored) + committed deletion hints (adjacency); overlap stacks naturally.
+- `src/extension.ts` — inject `DiffViewController` + `AttributionController` into `EditorProposalController`.
+- Test: editor E2E suite.
+
+---
+
+### Task 6: Per-author decoration types; recolor proposals to Claude
+
+**Files:**
+- Modify: `src/editorProposalController.ts` (20-27 decoration types; 80-106 renderEditor)
+- Test: editor E2E
+
+**Interfaces:**
+- Consumes: `decorationPlan` (already imported), proposal list.
+- Produces: four insertion decoration types (`human`/`claude`) keyed by author, deletion-hint types colored by author; proposals decorate with Claude colors.
+
+- [ ] **Step 1: Replace the two decoration types with per-author maps**
+
+In `src/editorProposalController.ts`, replace lines 20-27:
+
+```ts
+  private readonly insDeco: Record<"human" | "claude", vscode.TextEditorDecorationType> = {
+    human: vscode.window.createTextEditorDecorationType({
+      backgroundColor: "rgba(63,185,80,0.14)", borderColor: "#3fb950",
+      border: "0 0 2px 0", borderStyle: "solid",
+    }),
+    claude: vscode.window.createTextEditorDecorationType({
+      backgroundColor: "rgba(88,166,255,0.15)", borderColor: "#58a6ff",
+      border: "0 0 2px 0", borderStyle: "solid",
+    }),
+  };
+  private readonly delDeco: Record<"human" | "claude", vscode.TextEditorDecorationType> = {
+    human: vscode.window.createTextEditorDecorationType({ after: { color: "#f85149" } }),
+    claude: vscode.window.createTextEditorDecorationType({ after: { color: "#bc8cff" } }),
+  };
+```
+
+Update the constructor's disposables push (39-40) to dispose all four:
+
+```ts
+      this.insDeco.human, this.insDeco.claude, this.delDeco.human, this.delDeco.claude, this.lensEmitter,
+```
+
+- [ ] **Step 2: Recolor proposal decorations to Claude in `renderEditor`**
+
+Replace the body of `renderEditor` (81-106) so it groups ranges per author and applies the right decoration type. Proposals are Claude:
+
+```ts
+  private renderEditor(editor: vscode.TextEditor): void {
+    const doc = editor.document;
+    const clearAll = () => {
+      for (const a of ["human", "claude"] as const) {
+        editor.setDecorations(this.insDeco[a], []);
+        editor.setDecorations(this.delDeco[a], []);
+      }
+    };
+    if (doc.languageId !== "markdown") return clearAll();
+    const key = this.proposals.keyFor(doc);
+    const ins: Record<"human" | "claude", vscode.Range[]> = { human: [], claude: [] };
+    const del: Record<"human" | "claude", vscode.DecorationOptions[]> = { human: [], claude: [] };
+    // Pending proposals — always Claude (INV: proposals are agent-authored).
+    for (const v of this.proposals.listProposals(doc)) {
+      if (v.anchorStart === null || v.original === undefined || !this.proposals.isApplied(key, v.id)) continue;
+      const plan = decorationPlan(v.anchorStart, v.original, v.replacement);
+      for (const i of plan.insertions) ins.claude.push(new vscode.Range(doc.positionAt(i.start), doc.positionAt(i.end)));
+      for (const d of plan.deletions) {
+        del.claude.push({
+          range: new vscode.Range(doc.positionAt(d.at), doc.positionAt(d.at)),
+          renderOptions: { after: { contentText: ` ${d.text} `, textDecoration: "line-through" } },
+        });
+      }
+    }
+    // Committed changes-since-baseline are added in Task 7 (pushes into ins/del[author]).
+    this.decorateCommitted(editor, ins, del); // no-op stub until Task 7
+    for (const a of ["human", "claude"] as const) {
+      editor.setDecorations(this.insDeco[a], ins[a]);
+      editor.setDecorations(this.delDeco[a], del[a]);
+    }
+  }
+
+  /** Overridden in Task 7 to add committed (non-proposal) author-colored changes. */
+  private decorateCommitted(
+    _editor: vscode.TextEditor,
+    _ins: Record<"human" | "claude", vscode.Range[]>,
+    _del: Record<"human" | "claude", vscode.DecorationOptions[]>,
+  ): void {}
+```
+
+- [ ] **Step 3: Typecheck + build**
+
+Run: `npm run typecheck && npm run build`
+Expected: clean.
+
+- [ ] **Step 4: E2E — proposal shows Claude colors**
+
+Update/extend the editor E2E to assert a pending proposal decorates (smoke: the decoration types exist and apply; the harness asserts via the proposal-applied path used today). Run: `npm run test:e2e` → PASS.
+
+- [ ] **Step 5: Commit**
+
+```bash
+git add src/editorProposalController.ts test/e2e
+git commit -m "feat: per-author editor decoration types; proposals decorate in Claude blue/purple"
+```
+
+---
+
+### Task 7: Decorate committed changes-since-baseline by author (reverse INV-32)
+
+**Files:**
+- Modify: `src/editorProposalController.ts` — inject baseline + attribution; implement `decorateCommitted`; subscribe to baseline/attribution change events
+- Modify: `src/extension.ts` (line 122) — pass the two new deps
+- Test: editor E2E
+
+**Interfaces:**
+- Consumes: `DiffViewController.getBaseline(uri)` (diffViewController.ts:133) + `onDidChangeBaseline` (32); `AttributionController.spansFor(document)` (attributionController.ts:405); pure `wordEditHunks` + `authorAt` (trackChangesModel.ts).
+- Produces: committed insertions colored by `authorAt(currentOffset)`; committed deletion hints colored by adjacency (the paired insertion's author).
+
+- [ ] **Step 1: Inject the new dependencies**
+
+Change the constructor signature (38):
+
+```ts
+  constructor(
+    private readonly proposals: ProposalController,
+    private readonly diffView: import("./diffViewController").DiffViewController,
+    private readonly attribution: import("./attributionController").AttributionController,
+  ) {
+```
+
+Add subscriptions in the constructor disposables list:
+
+```ts
+      this.diffView.onDidChangeBaseline(({ uri }) => {
+        const ed = vscode.window.visibleTextEditors.find((e) => e.document.uri.toString() === uri);
+        if (ed) this.renderEditor(ed);
+      }),
+```
+
+In `src/extension.ts` (122) pass the deps (both already constructed above — `diffViewController` and `attributionController`):
+
+```ts
+  const editorProposalController = new EditorProposalController(
+    proposalController,
+    diffViewController,
+    attributionController,
+  );
+```
+
+(Confirm the local variable names in `extension.ts` for the diff-view + attribution controllers via `npm run typecheck`.)
+
+- [ ] **Step 2: Implement `decorateCommitted`**
+
+Replace the stub with the real implementation. It diffs baseline → current, attributes insertions, and colors deletion hints by adjacency:
+
+```ts
+  private decorateCommitted(
+    editor: vscode.TextEditor,
+    ins: Record<"human" | "claude", vscode.Range[]>,
+    del: Record<"human" | "claude", vscode.DecorationOptions[]>,
+  ): void {
+    const doc = editor.document;
+    const baseline = this.diffView.getBaseline(doc.uri.toString());
+    if (!baseline || baseline.reason === "pinned") return; // pin→clean: no committed marks
+    const current = doc.getText();
+    const spans = this.attribution.spansFor(doc); // AuthorSpan[] in current coords
+    const hunks = wordEditHunks(baseline.text, current); // start/end index INTO current
+    for (const h of hunks) {
+      // The replacement text occupies [h.start, h.start+replacement.length) in current.
+      // Re-derive the intra-hunk word diff to split ins vs del and find the author.
+      const original = baseline.text.slice(/* mapped */ 0, 0); // see note
+      const plan = decorationPlan(h.start, hunkOriginal(baseline.text, current, h), current.slice(h.start, h.end));
+      const author = authorAt(h.start, spans) ?? "human";
+      for (const i of plan.insertions) ins[author].push(new vscode.Range(doc.positionAt(i.start), doc.positionAt(i.end)));
+      for (const d of plan.deletions) {
+        del[author].push({
+          range: new vscode.Range(doc.positionAt(d.at), doc.positionAt(d.at)),
+          renderOptions: { after: { contentText: ` ${d.text} `, textDecoration: "line-through" } },
+        });
+      }
+    }
+  }
+```
+
+> **Implementation note for the engineer:** `wordEditHunks(baseline, current)` returns hunks whose `start`/`end` index into **current** with a `replacement` that is the *baseline* text for that span (see trackChangesModel.ts:241-264 — `replacement` accumulates `removed` parts = baseline text, `end` advances over them). So for a committed change, the *original* (baseline) text of a hunk is `h.replacement` and the *applied* (current) text is `current.slice(h.start, h.end)`. Call `decorationPlan(h.start, h.replacement, current.slice(h.start, h.end))` directly — drop the `hunkOriginal`/`original` placeholder lines above. Verify this against a unit test before wiring (Step 3).
+
+- [ ] **Step 3: Unit-test the hunk→plan derivation (pure)**
+
+Because the offset semantics are subtle, add a pure unit test in `test/trackChangesModel.test.ts` proving `wordEditHunks` + `decorationPlan` reconstruct a committed change. Add an exported helper if needed:
+
+```ts
+test("committed change: wordEditHunks replacement is baseline text; current slice is applied", () => {
+  const baseline = "the light mode";
+  const current = "the dark mode";
+  const [h] = wordEditHunks(baseline, current);
+  expect(h.replacement).toContain("light");          // baseline side
+  expect(current.slice(h.start, h.end)).toContain("dark"); // applied side
+  const plan = decorationPlan(h.start, h.replacement, current.slice(h.start, h.end));
+  expect(plan.insertions.length).toBeGreaterThan(0);
+  expect(plan.deletions.some((d) => d.text.includes("light"))).toBe(true);
+});
+```
+
+Run: `npm test -- trackChangesModel` → PASS. Then finalize `decorateCommitted` to use `h.replacement` + `current.slice(h.start, h.end)` (removing the placeholder lines).
+
+- [ ] **Step 4: Add the attribution-change refresh**
+
+Editor decorations must refresh when attribution changes (e.g. after a human edit). If `AttributionController` exposes a change event, subscribe to it and call `renderEditor`; otherwise refresh on `vscode.workspace.onDidChangeTextDocument` for the active editor (debounced like `scheduleApply`). Add to the constructor disposables:
+
+```ts
+      vscode.workspace.onDidChangeTextDocument((e) => {
+        const ed = vscode.window.activeTextEditor;
+        if (ed && e.document === ed.document) this.scheduleRender(ed);
+      }),
+```
+
+with a small `scheduleRender(ed)` that debounces `renderEditor(ed)` (mirror `scheduleApply`, 50ms).
+
+- [ ] **Step 5: Typecheck + build + E2E**
+
+Run: `npm run typecheck && npm run build && npm run test:e2e`
+Expected: clean. E2E: a committed human edit shows green underline; an accepted Claude replace shows blue underline + purple struck hint; a pinned baseline shows nothing.
+
+- [ ] **Step 6: Commit**
+
+```bash
+git add src/editorProposalController.ts src/extension.ts test/trackChangesModel.test.ts test/e2e
+git commit -m "feat: author-colored committed track changes inline in the editor (reverse INV-32)"
+```
+
+---
+
+### Task 8: Overlap — stacked marks where both authors touch one span
+
+**Files:**
+- Test: `test/trackChangesModel.test.ts` (preview overlap) + editor E2E (editor overlap)
+- Modify (if needed): `media/preview.css` (ensure `<ins>` nested in `<del>` and vice-versa render both marks)
+
+**Interfaces:**
+- Consumes: existing emitted classes.
+- Produces: a span both inserted (author A) and being deleted (author B) shows both A's underline and B's strikethrough.
+
+- [ ] **Step 1: Editor overlap — verify decorations stack**
+
+In the editor, an overlap arises when a committed insertion range (author A) is also covered by a pending proposal deletion (Claude). VS Code applies multiple decoration types to overlapping ranges, so the author-A insertion underline and the Claude deletion hint already coexist. Add an editor E2E asserting both decoration types are present on the overlap region. No code change expected; if the proposal's optimistic apply *replaces* the text (so A's range no longer exists), document that committed-insert + pending-delete overlap manifests as the proposal's struck original (Claude purple) adjacent to A's surviving insertion — assert that instead.
+
+- [ ] **Step 2: Preview overlap — CSS nesting**
+
+Add a CSS test fixture / assertion: a `<del class="cw-del-claude"><span class="cw-ins-human">…</span></del>` renders with both strikethrough (Claude purple) and underline (human green). Add to `media/preview.css` if the nested case collapses:
+
+```css
+.cw-del-claude .cw-ins-human, .cw-del-human .cw-ins-claude,
+.cw-del-claude .cw-by-human,  .cw-del-human .cw-by-claude { text-decoration: inherit; }
+```
+
+- [ ] **Step 3: Run tests**
+
+Run: `npm test && npm run test:e2e`
+Expected: PASS.
+
+- [ ] **Step 4: Commit**
+
+```bash
+git add media/preview.css test/trackChangesModel.test.ts test/e2e
+git commit -m "feat: overlap renders stacked author marks (insert + delete) in both panes"
+```
+
+---
+
+### Task 9: Sweep retired markup + full suite + manual smoke
+
+**Files:**
+- `test/**`, `media/preview.css`, any code referencing retired classes.
+
+- [ ] **Step 1: Grep for retired classes/behaviors**
+
+Run:
+```bash
+grep -rn "cw-add\b\|cw-del\b\|diffEditor.insertedTextBackground\|gitDecoration.deletedResourceForeground" src/ media/ test/
+```
+Reconcile every hit: `cw-add`/`cw-del` (without `-author`) only legitimately remain in `#cw-summary`. The editor's old semantic ThemeColors are gone (replaced in Task 6).
+
+- [ ] **Step 2: Full unit + typecheck + build**
+
+Run: `npm run typecheck && npm test && npm run build`
+Expected: all green.
+
+- [ ] **Step 3: Full E2E**
+
+Run: `npm run test:e2e`
+Expected: all green (update any remaining old-class assertions).
+
+- [ ] **Step 4: Manual smoke (both panes)**
+
+Open the sandbox doc, make a human edit + ask Claude to edit. Confirm: editor shows green/blue underlines + purple/red struck hints; preview shows the same; unchanged text plain in both; pin baseline → both panes clean; accept a proposal → it recolors as a committed change until re-pin.
+
+- [ ] **Step 5: Commit + open PR**
+
+```bash
+git add -A
+git commit -m "chore: sweep retired semantic-diff markup; full suite green"
+```
+
+---
+
+## Self-Review
+
+**1. Spec coverage:**
+- Model (style=op, color=author) → Tasks 1-4 (preview), 6-7 (editor). ✓
+- Human green/red + Claude blue/purple, exact hexes → Global Constraints + Task 4 CSS + Task 6 decorations. ✓
+- Both panes → Phase 1 (preview) + Phase 2 (editor). ✓
+- Reverses INV-32 → Task 7. ✓
+- Both overlap directions, stacked → Task 8. ✓
+- Color = changed-since-baseline; pin→clean preserved → Task 2 (unchanged plain), Task 7 (`reason === "pinned"` guard). ✓
+- Deletion authorship = adjacency heuristic; standalone neutral → Task 1 (`wordDiffByAuthor` cluster author + `cw-del-none`), Task 7 (`authorAt(h.start)`). ✓
+- Components: trackChangesModel, editorProposalController, attributionController (read via spansFor), preview.css, extension.ts wiring → covered. ✓
+- Testing: unit (pure render), editor decorations (E2E), host E2E, retire old assertions → Tasks 1-3, 5, 7-9. ✓
+
+**2. Placeholder scan:** Task 7 Step 2 intentionally shows a *wrong* placeholder (`hunkOriginal`/`original`) and immediately corrects it in the implementation note + Step 3 unit test that pins the exact offsets — this is a guided derivation, not an unresolved placeholder. No other TBD/TODO.
+
+**3. Type consistency:** `AuthorKind` (`"claude" | "human"`) used throughout; decoration maps keyed `"human" | "claude"`; `authorAt` returns `AuthorKind | null` → `?? "human"`/`?? "none"` at every call site; `ProposalView.author?: AuthorKind`; `wordDiffByAuthor(beforeRaw, afterRaw, afterStart, spans)` signature matches both its test and its `renderReview` call site.
+
+## Execution Handoff
+
+(Filled in after you review.)
@@ -0,0 +1,526 @@
+# #70 — INV-5 Reject-After-Interior-Edit Gap Implementation Plan
+
+> **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.
+
+**Goal:** `✗ Reject` on an optimistically-applied proposal restores the retained original (INV-5) even after the writer typed inside the pending range — and when it genuinely cannot locate the applied span, it fails loudly instead of reporting success while leaving the proposed text behind.
+
+**Architecture:** Issue #70's fix direction **(a)** with an honest **(b)** fallback. A new per-doc `appliedSpans: Map<proposalId, OffsetRange>` becomes the real F12 applied-range bookkeeping: written at optimistic-apply time, re-synced whenever the exact anchor resolves at `renderAll`, shifted through interior-safe buffer edits (`shift()`), and **deleted (distrusted)** when an edit straddles a span boundary (e.g. a whole-buffer external replace — a shifted range clamps to garbage there, and restoring at a guessed offset would corrupt the doc, INV-11's "never applied by guess"). `revertInPlace` then reverts via exact resolve → `appliedSpans` fallback → hard failure (warning, proposal kept, `false` returned). Direction (c) (fuzzy resolve) is rejected: it guesses. `rejectAll` orders by the same fallback and reports skips like `acceptAllProposals` does.
+
+**Tech Stack:** TypeScript VS Code extension; `npm test` (vitest unit), `npm run test:e2e` (@vscode/test-electron host suite — `ProposalController` is vscode-coupled, so coverage is host E2E, matching every existing F12 test).
+
+## Global Constraints
+
+- INV-5: "Reject restores the retained original exactly" — the contract this fix enforces; `revertInPlace`'s doc comment already promises "reverts the whole block regardless of any in-place edits the human made".
+- INV-11: anchors are immutable for a proposal's life; stale text is flagged, **never applied by guess** — the fallback must only use a span whose tracking is provably continuous (no fuzzy resolve, no boundary-straddled ranges).
+- INV-12: accept/reject stay human-only gestures; no new mutation paths.
+- Keep `✓ Keep` (`finalizeInPlace`) behavior byte-identical — its lookup-by-id bypass of resolution is correct by design (issue "Contrast" note).
+- Existing test suites must stay green: 265+ unit, full host E2E (notably `proposals.test.ts` INV-11 stale-accept, `f12InlineDiff.test.ts`, `fullLoop.test.ts`).
+- Wiggleverse hygiene: no inline trailing comments in CLI blocks; commits cite the issue and carry the `Co-Authored-By` trailer.
+
+---
+
+### Task 1: `appliedSpans` bookkeeping + `revertInPlace` tracked-range fallback (the INV-5 repro)
+
+**Files:**
+- Modify: `src/proposalController.ts` (DocState ~line 35, `ensureState` ~127, `optimisticApply` ~336, `finalizeInPlace` ~417, `revertInPlace` ~434, `renderAll` ~495, `onDidChange` ~529)
+- Test: `test/e2e/suite/f12InlineDiff.test.ts` (new suite appended)
+
+**Interfaces:**
+- Consumes: `shift(range, edit)` / `resolve(text, fp)` from `src/anchorer.ts` (existing); `DocState.applied: Set<string>` (existing).
+- Produces: `DocState.appliedSpans: Map<string, OffsetRange>` (private bookkeeping; Tasks 2–3 rely on its distrust + ordering semantics); `revertInPlace(docPath, proposalId, opts?: { silent?: boolean }): Promise<boolean>` — signature gains the optional `opts` used by Task 3.
+
+- [ ] **Step 1: Write the failing E2E test — the issue's repro sketch**
+
+Append to `test/e2e/suite/f12InlineDiff.test.ts`:
+
+```typescript
+// #70 (INV-5): rejecting a proposal AFTER typing inside its pending range must
+// still restore the retained original. The exact-substring anchor is orphaned by
+// the interior tweak (INV-1/INV-11) — the revert falls back to the F12
+// applied-range bookkeeping (appliedSpans), which shift-tracks the span through
+// interior edits.
+suite("F12 inline diff — #70 reject after interior edit (INV-5)", () => {
+  test("revertInPlace restores the original after a tweak inside the pending range", async () => {
+    const ORIGINAL = "The quick brown fox jumps.";
+    const { doc } = await freshDoc("docs/s70-reject-tweaked.md", `# T\n\n${ORIGINAL}\n`);
+    const api = await getApi();
+    const p = api.proposalController;
+    const docKey = p.keyFor(doc);
+    const fp = { text: ORIGINAL, before: "", after: "", lineHint: 2 };
+    const id = await p.propose(doc, fp, "The rapid brown fox jumps.",
+      { kind: "agent", id: "claude", agent: { sdk: "x", model: "m", sessionId: "s" } }, { granularity: "block" });
+    await p.optimisticApply(doc, id!);
+    await settle();
+    assert.ok(doc.getText().includes("The rapid brown fox jumps."), "optimistically applied");
+    // Human types INSIDE the pending range → orphans the exact anchor (INV-11).
+    const at = doc.getText().indexOf("rapid");
+    const we = new vscode.WorkspaceEdit();
+    we.replace(doc.uri, new vscode.Range(doc.positionAt(at), doc.positionAt(at + "rapid".length)), "RAPID-ish");
+    assert.ok(await vscode.workspace.applyEdit(we), "interior tweak applied");
+    await settle();
+    assert.strictEqual(p.listProposals(doc).find((v) => v.id === id)!.anchorStart, null, "anchor orphaned by the tweak");
+    // ✗ Reject: must restore the retained original EXACTLY (INV-5), not skip.
+    assert.ok(await p.revertInPlace(docKey, id!), "reject reports success");
+    await settle();
+    assert.strictEqual(doc.getText(), `# T\n\n${ORIGINAL}\n`, "original restored exactly (INV-5)");
+    assert.strictEqual(p.listProposals(doc).length, 0, "proposal cleared");
+  });
+
+  test("rejectByIdInPlace routes the tweaked-applied case the same way", async () => {
+    const ORIGINAL = "A steady closing line.";
+    const { doc } = await freshDoc("docs/s70-reject-route.md", `# T\n\n${ORIGINAL}\n`);
+    const api = await getApi();
+    const p = api.proposalController;
+    const docKey = p.keyFor(doc);
+    const fp = { text: ORIGINAL, before: "", after: "", lineHint: 2 };
+    const id = await p.propose(doc, fp, "A wobbly closing line.",
+      { kind: "agent", id: "claude", agent: { sdk: "x", model: "m", sessionId: "s" } }, { granularity: "block" });
+    await p.optimisticApply(doc, id!);
+    await settle();
+    const at = doc.getText().indexOf("wobbly");
+    const we = new vscode.WorkspaceEdit();
+    we.replace(doc.uri, new vscode.Range(doc.positionAt(at), doc.positionAt(at)), "very-");
+    assert.ok(await vscode.workspace.applyEdit(we), "interior insertion applied");
+    await settle();
+    assert.ok(await p.rejectByIdInPlace(docKey, id!), "gesture-level reject succeeds");
+    await settle();
+    assert.strictEqual(doc.getText(), `# T\n\n${ORIGINAL}\n`, "original restored exactly (INV-5)");
+  });
+});
+```
+
+- [ ] **Step 2: Run the new suite to verify it fails**
+
+```bash
+npm run test:e2e 2>&1 | grep -A 3 "#70"
+```
+
+Expected: both tests FAIL — first on `doc.getText()` still holding the tweaked replacement (`RAPID-ish`) after a `true` return (the silent-skip bug), i.e. the "original restored exactly" assertion.
+
+- [ ] **Step 3: Implement `appliedSpans` + the fallback**
+
+In `src/proposalController.ts`:
+
+(1) `DocState` gains the map (after `applied`):
+
+```typescript
+  /** ids already optimistically applied to the buffer (so the trigger doesn't re-apply). */
+  applied: Set<string>;
+  /**
+   * #70 (INV-5): the F12 applied-range bookkeeping — proposal id -> the applied
+   * span's CURRENT buffer range. Written at optimistic-apply, re-synced whenever
+   * the exact anchor resolves (renderAll), shifted through interior-safe edits,
+   * and DELETED (distrusted) when an edit straddles a span boundary — a clamped
+   * range is a guess, and stale text is never applied by guess (INV-11). Lets
+   * Reject restore the original after the human types INSIDE the pending range
+   * (which orphans the exact-substring anchor).
+   */
+  appliedSpans: Map<string, OffsetRange>;
+```
+
+(2) `ensureState` initializes it: `appliedSpans: new Map(),` after `applied: new Set(),`.
+
+(3) `optimisticApply` records the span where it already re-anchors (after `state.applied.add(proposalId);` and the `appliedStart`/`appliedEnd` computation — move the `add` down beside it or record right after the existing lines):
+
+```typescript
+    state.applied.add(proposalId);
+    // Re-anchor to the applied text now in the buffer (its start is unchanged; its
+    // end shifts by the net length delta of the replacement).
+    const appliedStart = resolved.start;
+    const appliedEnd = appliedStart + proposal.replacement.length;
+    state.appliedSpans.set(proposalId, { start: appliedStart, end: appliedEnd });
+```
+
+(4) `renderAll`'s resolved branch re-syncs (covers the prior-session-applied reload, where the in-memory map starts empty):
+
+```typescript
+      } else {
+        this.recordProposal(state, proposal, resolved, true);
+        if (state.applied.has(proposal.id)) state.appliedSpans.set(proposal.id, resolved);
+      }
+```
+
+(5) `onDidChange` maintains it — interior-safe edits shift, boundary-straddling edits distrust. Replace the method body:
+
+```typescript
+  private onDidChange(e: vscode.TextDocumentChangeEvent): void {
+    const state = this.docs.get(this.keyOf(e.document));
+    if (!state || (state.live.size === 0 && state.appliedSpans.size === 0)) return;
+    for (const change of e.contentChanges) {
+      const edit = { start: change.rangeOffset, end: change.rangeOffset + change.rangeLength, newLength: change.text.length };
+      for (const [id, range] of state.live) state.live.set(id, shift(range, edit));
+      // #70: an edit fully inside a tracked applied span (or fully outside it)
+      // keeps the span meaningful; one straddling a boundary — e.g. a whole-buffer
+      // external replace — makes the shifted range a clamped guess, so the span
+      // is distrusted (deleted) rather than reverted-to by guess (INV-11).
+      for (const [id, range] of state.appliedSpans) {
+        const outside = edit.end <= range.start || edit.start >= range.end;
+        const inside = edit.start >= range.start && edit.end <= range.end;
+        if (outside || inside) state.appliedSpans.set(id, shift(range, edit));
+        else state.appliedSpans.delete(id);
+      }
+    }
+  }
+```
+
+(6) `revertInPlace` — the fix itself. Replace the method (keep its position; doc comment updated to describe the layered lookup; `opts` is consumed by Task 3):
+
+```typescript
+  /**
+   * F12/#64 (INV-51): REJECT an optimistically-applied proposal — replace its live
+   * applied span with the stored `original`, then clear it. Reverts the whole block
+   * regardless of any in-place edits the human made to the inserted text: when an
+   * interior tweak has orphaned the exact anchor (INV-11), the revert falls back to
+   * the shift-tracked applied span (#70, INV-5). When neither locates the span
+   * (e.g. a boundary-straddling external rewrite distrusted it), the reject FAILS
+   * loudly — warning shown, proposal kept, `false` returned — instead of silently
+   * leaving the proposed text in the buffer. Undo (or ✓ Keep) remains the way out.
+   */
+  async revertInPlace(docPath: string, proposalId: string, opts?: { silent?: boolean }): Promise<boolean> {
+    const hit = this.byId(docPath, proposalId);
+    if (!hit) return false;
+    const document = this.openDoc(hit.state);
+    if (!document) return false;
+    // Never optimistically applied → nothing of ours is in the buffer; clearing
+    // the record IS the reject (the legacy pending-only path).
+    if (hit.proposal.original === undefined) {
+      this.store.update(docPath, (a) => removeProposal(a, proposalId));
+      hit.state.applied.delete(proposalId);
+      hit.state.appliedSpans.delete(proposalId);
+      this.renderAll(document);
+      return true;
+    }
+    const fp = hit.state.artifact.anchors[hit.proposal.anchorId]?.fingerprint;
+    const resolved = fp ? resolve(document.getText(), fp) : "orphaned";
+    const span = resolved !== "orphaned" ? resolved : hit.state.appliedSpans.get(proposalId);
+    if (!span) {
+      if (!opts?.silent) {
+        void vscode.window.showWarningMessage(
+          "Cowriting: this proposal's applied text can't be located (it changed or moved) — undo your edits, or Keep it to leave the buffer as-is (it is never reverted by guess).",
+        );
+      }
+      return false;
+    }
+    const we = new vscode.WorkspaceEdit();
+    we.replace(
+      document.uri,
+      new vscode.Range(document.positionAt(span.start), document.positionAt(span.end)),
+      hit.proposal.original,
+    );
+    if (!(await vscode.workspace.applyEdit(we))) return false;
+    this.store.update(docPath, (a) => removeProposal(a, proposalId));
+    hit.state.applied.delete(proposalId);
+    hit.state.appliedSpans.delete(proposalId);
+    this.renderAll(document);
+    return true;
+  }
+```
+
+(7) `finalizeInPlace` cleans the map beside its existing `applied.delete`:
+
+```typescript
+    hit.state.applied.delete(proposalId);
+    hit.state.appliedSpans.delete(proposalId);
+```
+
+- [ ] **Step 4: Typecheck + unit tests + the new E2E**
+
+```bash
+npm run typecheck && npm test
+npm run test:e2e
+```
+
+Expected: typecheck clean; all unit tests PASS (no unit test touches `ProposalController`); host E2E PASS including the two new #70 tests and all pre-existing F12/proposals/fullLoop suites.
+
+- [ ] **Step 5: Commit**
+
+```bash
+git add src/proposalController.ts test/e2e/suite/f12InlineDiff.test.ts
+git commit -m "fix(#70): reject after an interior tweak restores the original via tracked applied span (INV-5)
+
+Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>"
+```
+
+---
+
+### Task 2: The honest hard-failure path (fix direction (b) for the residual orphan)
+
+**Files:**
+- Modify: `src/proposalController.ts` (only if Step 2 exposes a gap — the Task 1 code already implements the path)
+- Test: `test/e2e/suite/f12InlineDiff.test.ts` (extend the #70 suite)
+
+**Interfaces:**
+- Consumes: Task 1's `appliedSpans` distrust semantics and `revertInPlace` failure contract.
+- Produces: nothing new — locks the contract with a regression test.
+
+- [ ] **Step 1: Write the failing/locking E2E test**
+
+Append inside the `#70` suite from Task 1:
+
+> **Execution note:** the first cut of this test used a whole-buffer rewrite as the
+> distrusting edit and FAILED — VS Code minimizes workspace edits before change
+> events fire, so a rewrite sharing a prefix/suffix with the old text decomposes
+> into interior hunks the span legitimately survives (and the revert then correctly
+> restores the original in place). The distrusting edit must straddle a span
+> boundary even after minimization — a deletion from outside the span into its
+> interior, as below.
+
+```typescript
+  // Fix direction (b): when the tracked span itself is distrusted (an edit
+  // straddled its boundary — here a deletion running from the heading into the
+  // span's interior), the reject FAILS honestly: warning path, proposal kept,
+  // buffer untouched. Silent-skip-and-report-success (the #70 bug) must not come
+  // back; reverting at a clamped guessed offset (INV-11) must not either.
+  // (A boundary-straddling DELETION is used because VS Code minimizes workspace
+  // edits before change events fire — a wholesale rewrite sharing a prefix/suffix
+  // decomposes into interior hunks the span legitimately survives.)
+  test("reject fails loudly — proposal kept, buffer untouched — when the span is distrusted", async () => {
+    const ORIGINAL = "Sentence one stands here.";
+    const { doc } = await freshDoc("docs/s70-reject-distrust.md", `# T\n\n${ORIGINAL}\n`);
+    const api = await getApi();
+    const p = api.proposalController;
+    const docKey = p.keyFor(doc);
+    const fp = { text: ORIGINAL, before: "", after: "", lineHint: 2 };
+    const id = await p.propose(doc, fp, "Sentence ONE stands here.",
+      { kind: "agent", id: "claude", agent: { sdk: "x", model: "m", sessionId: "s" } }, { granularity: "block" });
+    await p.optimisticApply(doc, id!);
+    await settle();
+    // Delete from inside the heading through the middle of the applied span:
+    // the edit straddles the span's start boundary → the tracked span is
+    // distrusted AND the exact anchor no longer resolves.
+    const mid = doc.getText().indexOf("ONE");
+    const we = new vscode.WorkspaceEdit();
+    we.replace(doc.uri, new vscode.Range(doc.positionAt(2), doc.positionAt(mid + 1)), "");
+    assert.ok(await vscode.workspace.applyEdit(we), "boundary-straddling deletion applied");
+    await settle();
+    const before = doc.getText();
+    assert.strictEqual(await p.revertInPlace(docKey, id!), false, "reject reports failure (no silent success)");
+    await settle();
+    assert.strictEqual(doc.getText(), before, "buffer untouched — never reverted by guess (INV-11)");
+    assert.ok(p.listProposals(doc).some((v) => v.id === id), "proposal kept (not silently dropped)");
+    // Cleanup for later tests: the never-locatable husk is discarded via the
+    // plain record-only reject.
+    assert.strictEqual(p.rejectById(docKey, id!), true);
+  });
+```
+
+- [ ] **Step 2: Run it**
+
+```bash
+npm run test:e2e 2>&1 | grep -B 2 -A 5 "distrusted"
+```
+
+Expected: PASS if Task 1's implementation is complete (this is the locking regression); if it FAILS, the failure pinpoints which leg (distrust bookkeeping vs. failure return vs. proposal retention) is wrong — fix `src/proposalController.ts` accordingly, not the test.
+
+- [ ] **Step 3: Commit**
+
+```bash
+git add test/e2e/suite/f12InlineDiff.test.ts src/proposalController.ts
+git commit -m "test(#70): lock the honest hard-failure reject path (distrusted span → warn, keep, false)
+
+Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>"
+```
+
+---
+
+### Task 3: `rejectAll` parity — tweaked proposals revert too, skips are reported
+
+**Files:**
+- Modify: `src/proposalController.ts` (`rejectAll` ~line 461), `src/extension.ts` (`cowriting.rejectAllProposals` handler ~line 220)
+- Test: `test/e2e/suite/f12InlineDiff.test.ts` (extend the #70 suite)
+
+**Interfaces:**
+- Consumes: Task 1's `revertInPlace(docPath, id, opts?: { silent?: boolean })` and `appliedSpans`.
+- Produces: `rejectAll(document): Promise<{ reverted: number; skipped: number }>` — additive `skipped` field (existing `{ reverted }` destructurings stay valid).
+
+- [ ] **Step 1: Write the failing E2E test**
+
+Append inside the `#70` suite:
+
+```typescript
+  // rejectAll must revert a tweaked (orphaned-anchor) proposal via the same
+  // tracked-span fallback, ordered descending by that span so earlier reverts
+  // never shift later ones — and must count what it could not revert.
+  test("rejectAll reverts tweaked proposals via the tracked span and reports skips", async () => {
+    const { doc } = await freshDoc("docs/s70-rejall-tweak.md", "# T\n\nOne aaa.\n\nTwo bbb.\n");
+    const api = await getApi();
+    const p = api.proposalController;
+    const editFlow = api.editFlow;
+    editFlow.setEditTurnForTest(async () => ({ replacement: "# T\n\nOne AAA.\n\nTwo BBB.\n", model: "m", sessionId: "s" }));
+    const ids = await editFlow.runEditAndPropose(doc, { kind: "document" }, "up");
+    await settle();
+    for (const id of ids) await p.optimisticApply(doc, id);
+    await settle();
+    // Tweak INSIDE the first applied block → its exact anchor orphans.
+    const at = doc.getText().indexOf("AAA");
+    const we = new vscode.WorkspaceEdit();
+    we.replace(doc.uri, new vscode.Range(doc.positionAt(at), doc.positionAt(at + 3)), "AAAZ");
+    assert.ok(await vscode.workspace.applyEdit(we), "interior tweak applied");
+    await settle();
+    const { reverted, skipped } = await p.rejectAll(doc);
+    await settle();
+    assert.strictEqual(reverted, ids.length, "ALL proposals reverted, tweaked one included");
+    assert.strictEqual(skipped, 0, "nothing skipped");
+    assert.strictEqual(doc.getText(), "# T\n\nOne aaa.\n\nTwo bbb.\n", "document fully restored (INV-5)");
+    assert.strictEqual(p.listProposals(doc).length, 0, "all cleared");
+  });
+```
+
+- [ ] **Step 2: Run it to verify the current gap**
+
+```bash
+npm run test:e2e 2>&1 | grep -B 2 -A 5 "rejectAll reverts tweaked"
+```
+
+Expected: FAIL — today the tweaked proposal sorts as an orphan (`start: -1`, reverted last), and after the first successful revert's `renderAll` the old code silently removes it without restoring (`document fully restored` assertion fails), and `skipped` does not exist on the return type (typecheck catches first — that is the same failure).
+
+- [ ] **Step 3: Implement — order by the fallback span, count skips, report them**
+
+Replace `rejectAll` in `src/proposalController.ts`:
+
+```typescript
+  /**
+   * F12/#64 (INV-53): reject EVERY pending proposal on a document — revert each in
+   * DESCENDING span order (so an earlier revert never shifts a later one's
+   * offsets), symmetric with #46's accept-all. #70: a tweaked proposal whose exact
+   * anchor is orphaned orders (and reverts) by its shift-tracked applied span;
+   * one with no locatable span is SKIPPED — counted, never guessed (INV-11) —
+   * and the per-proposal warning is suppressed in favor of the batch report.
+   */
+  async rejectAll(document: vscode.TextDocument): Promise<{ reverted: number; skipped: number }> {
+    if (!this.isTracked(document)) return { reverted: 0, skipped: 0 };
+    const docPath = this.keyOf(document);
+    const state = this.ensureState(document);
+    state.artifact = this.store.load(docPath) ?? emptyArtifact(docPath);
+    const text = document.getText();
+    const ordered = state.artifact.proposals
+      .map((p) => {
+        const fp = state.artifact.anchors[p.anchorId]?.fingerprint;
+        const r = fp ? resolve(text, fp) : "orphaned";
+        const start = r !== "orphaned" ? r.start : state.appliedSpans.get(p.id)?.start ?? -1;
+        return { id: p.id, start };
+      })
+      .sort((a, b) => b.start - a.start);
+    let reverted = 0;
+    let skipped = 0;
+    for (const it of ordered) {
+      if (await this.revertInPlace(docPath, it.id, { silent: true })) reverted++;
+      else skipped++;
+    }
+    return { reverted, skipped };
+  }
+```
+
+And in `src/extension.ts`, the `cowriting.rejectAllProposals` handler mirrors accept-all's skip note:
+
+```typescript
+      const { reverted, skipped } = await proposalController.rejectAll(doc);
+      if (reverted === 0 && skipped === 0) return;
+      const skipNote = skipped > 0 ? `, ${skipped} skipped (applied text not locatable — undo your edits or Keep)` : "";
+      void vscode.window.showInformationMessage(
+        `Cowriting: rejected ${reverted} proposal${reverted === 1 ? "" : "s"}${skipNote}.`,
+      );
+```
+
+- [ ] **Step 4: Typecheck + full test run**
+
+```bash
+npm run typecheck && npm test
+npm run test:e2e
+```
+
+Expected: all PASS — including the pre-existing `rejectAll` E2E tests (`f12InlineDiff` "rejectAll reverts every pending proposal", "control parity", `proposals.test.ts` cleanup calls), whose `{ reverted }` destructuring is unaffected by the additive field.
+
+- [ ] **Step 5: Commit**
+
+```bash
+git add src/proposalController.ts src/extension.ts test/e2e/suite/f12InlineDiff.test.ts
+git commit -m "fix(#70): rejectAll reverts tweaked proposals via the tracked span; skips are counted and reported
+
+Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>"
+```
+
+---
+
+### Task 4: Close the loop — fullLoop E2E rejects a tweaked proposal (the deliberate avoidance falls)
+
+**Files:**
+- Modify: `test/e2e/suite/fullLoop.test.ts` (the PUC-2 reject section, ~line 144)
+
+**Interfaces:**
+- Consumes: the shipped Task 1 behavior; existing fullLoop fixtures (`rejectId`, `REJECT_REPLACEMENT`, `ORIG_REJECT`).
+- Produces: nothing — upgrades the flagship E2E so the gap can't silently regress.
+
+- [ ] **Step 1: Tweak inside the reject proposal's range before rejecting**
+
+In `test/e2e/suite/fullLoop.test.ts`, immediately before the existing line
+`assert.ok(await api.proposalController.revertInPlace(docKey, rejectId), "reject reverts in place");`,
+insert:
+
+```typescript
+    // #70 (INV-5): the reject leg now ALSO takes an interior tweak first — the
+    // original deliberately rejected only an un-tweaked proposal because the
+    // pre-fix revert silently skipped an orphaned anchor. The tweak orphans the
+    // exact anchor; the revert must restore ORIG_REJECT via the tracked span.
+    const rejAt = doc.getText().indexOf(REJECT_REPLACEMENT);
+    assert.ok(rejAt >= 0, "reject proposal's applied text present");
+    const rejTweak = new vscode.WorkspaceEdit();
+    rejTweak.replace(doc.uri, new vscode.Range(doc.positionAt(rejAt + 2), doc.positionAt(rejAt + 2)), "x");
+    assert.ok(await vscode.workspace.applyEdit(rejTweak), "human tweak inside the reject proposal's range");
+    await settle();
+    assert.strictEqual(
+      api.proposalController.listProposals(doc).find((v) => v.id === rejectId)!.anchorStart,
+      null,
+      "tweak orphans the reject proposal's exact anchor (INV-11)",
+    );
+```
+
+(The existing assertions right after — proposal cleared, `ORIG_REJECT` restored, `REJECT_REPLACEMENT` gone — now verify the #70 path. If `REJECT_REPLACEMENT` is a multi-word string whose slice at `+2` splits a word the later `!doc.getText().includes(REJECT_REPLACEMENT)` assertion still holds; the `includes(ORIG_REJECT)` restore assertion is the INV-5 check.)
+
+- [ ] **Step 2: Run the fullLoop suite**
+
+```bash
+npm run test:e2e 2>&1 | grep -B 2 -A 8 "fullLoop\|full-loop\|full loop"
+```
+
+Expected: PASS end-to-end (would FAIL on unfixed code: the buffer would keep the tweaked `REJECT_REPLACEMENT`, so `includes(ORIG_REJECT)` fails).
+
+- [ ] **Step 3: Full verification sweep**
+
+```bash
+npm run typecheck && npm test && npm run test:e2e
+```
+
+Expected: everything green.
+
+- [ ] **Step 4: Commit**
+
+```bash
+git add test/e2e/suite/fullLoop.test.ts
+git commit -m "test(#70): fullLoop reject leg now tweaks inside the pending range first (INV-5 end-to-end)
+
+Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>"
+```
+
+---
+
+## Post-execution review wave (not in the original tasks)
+
+A high-effort multi-agent branch review after Task 4 confirmed 10 findings; all
+but one were fixed in a follow-up commit on this branch ("harden the
+tracked-span revert per branch review"): pure `shiftTracked` in `anchorer.ts`
+(+11 unit tests; insertion-at-span-start lands before the span), tracked spans
+cleared on document close, rebuild-only resync at `renderAll`, `guard.isReadOnly`
+on `revertInPlace`/`finalizeInPlace`, a "Discard proposal (leave text)" action on
+the hard-fail warning, CodeLens pair anchored at the tracked span (+1 E2E),
+QuickPick batch menu routed through the reporting commands, and one
+`clearProposal` helper (also fixes `reject()`'s stale `applied`/`appliedSpans`).
+The remaining finding — `resolve()` accepts a single exact occurrence without a
+context check, so a duplicated block can hijack any anchor consumer — pre-dates
+this fix and is filed as its own issue.
+
+## Self-review notes
+
+- **Issue coverage:** direction (a) → Task 1 (`appliedSpans` fallback); direction (b) → Task 2 (hard failure, warn, keep); direction (c) explicitly rejected (guessing violates INV-11). The issue's repro sketch is Task 1's test verbatim; the "full-loop E2E deliberately rejects only an un-tweaked proposal" note is retired by Task 4; `rejectAll` (same `revertInPlace` seam) is covered by Task 3.
+- **Type consistency:** `revertInPlace(docPath, proposalId, opts?: { silent?: boolean })` defined in Task 1, consumed in Task 3; `rejectAll` returns `{ reverted, skipped }` (additive); `appliedSpans: Map<string, OffsetRange>` used in Tasks 1 and 3.
+- **Contrast guard:** `finalizeInPlace` untouched except symmetric `appliedSpans.delete` cleanup — the Keep bypass stays.
@@ -0,0 +1,342 @@
+# #71 Surface Polish Batch Implementation Plan
+
+> **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.
+
+**Goal:** Land the six batched Minor findings from the native-surfaces migration's final whole-branch review (issue #71) — misleading `pin()` warning, orphaned command declarations, a wrong when-clause key, README F3/banner drift, an ungated palette entry, and an uncleared debounce timer.
+
+**Architecture:** Six independent point fixes across `src/diffViewController.ts`, `src/gitBaseline.ts`, `package.json`, and `README.md`. No new modules, no model/persistence changes, no new invariants. Verification is the existing suites (all six items are copy/declaration/cleanup changes with no unit-testable pure surface — the unit suite is vscode-free and these all live at the vscode boundary; E2E cannot observe warning-message text or palette visibility).
+
+**Tech Stack:** VS Code extension (TypeScript, esbuild CJS bundle), vitest unit suite (`npm test`), `@vscode/test-electron` host E2E (`npm run test:e2e`).
+
+## Global Constraints
+
+- Work happens on branch `worktree-s71-surface-polish` in the worktree at `.claude/worktrees/s71-surface-polish` (canonical checkout is occupied by concurrent session 0062 — never touch it).
+- Commits cite #71, read like surrounding history, and carry the `Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>` trailer.
+- Warning copy for the not-coediting case is exactly `"Run ✦ Coedit this Document with Claude first."` — the string `src/threadController.ts:181` already uses. Do not coin a variant.
+- No inline trailing comments on shell commands (wgl `no-inline-cli-comments`).
+- Final gate before PR: `npm test` (unit), `npm run typecheck`, `npm run build`, `npm run test:e2e` all green.
+
+---
+
+### Task 1: `pin()` explicit not-coediting branch (item 1)
+
+**Files:**
+- Modify: `src/diffViewController.ts:145-155`
+
+**Interfaces:**
+- Consumes: `this.modes: Map<string, "head" | "snapshot">` (set only by `establish()`).
+- Produces: nothing new — same `pin(document): void` signature; only the warning copy forks.
+
+`pin()` currently emits "this document is git-tracked — commit to advance the baseline" for **any** non-snapshot mode, including `undefined` (a document that never established, i.e. is not being coedited) — a wrong message for that case.
+
+- [ ] **Step 1: Fork the guard on `undefined`**
+
+Replace lines 145–155 (the whole `pin` method) with:
+
+```ts
+  pin(document: vscode.TextDocument): void {
+    if (!this.isDiffable(document)) return;
+    const key = this.uriKey(document);
+    const mode = this.modes.get(key);
+    if (mode === undefined) {
+      // never established — the document is not being coedited (#71 item 1)
+      void vscode.window.showWarningMessage("Run ✦ Coedit this Document with Claude first.");
+      return;
+    }
+    if (mode !== "snapshot") {
+      void vscode.window.showWarningMessage(
+        "Cowriting: this document is git-tracked — commit to advance the baseline.",
+      );
+      return;
+    }
+    this.capture(document, "pinned");
+  }
+```
+
+Keep the existing doc comment above the method unchanged.
+
+- [ ] **Step 2: Verify**
+
+Run: `npm run typecheck`
+Expected: clean exit.
+
+- [ ] **Step 3: Commit**
+
+```bash
+git add src/diffViewController.ts
+git commit -m "fix: markReviewed on a never-established doc says 'coedit first', not 'git-tracked' (#71 item 1)"
+```
+
+---
+
+### Task 2: delete orphaned `acceptProposal` / `rejectProposal` declarations (item 2)
+
+**Files:**
+- Modify: `package.json` (two `contributes.commands` entries ~85-93, two `contributes.menus.commandPalette` entries ~180-187)
+
+**Interfaces:** none — pure declaration removal. `test/e2e/suite-no-workspace/noWorkspace.test.ts:42-43` already asserts these commands are **not registered** at runtime; nothing registers them in `src/`.
+
+- [ ] **Step 1: Delete the two command declarations**
+
+In `contributes.commands`, delete both objects:
+
+```json
+      {
+        "command": "cowriting.acceptProposal",
+        "title": "✓ Accept Proposal",
+        "category": "Cowriting"
+      },
+      {
+        "command": "cowriting.rejectProposal",
+        "title": "✗ Reject Proposal",
+        "category": "Cowriting"
+      },
+```
+
+- [ ] **Step 2: Delete their palette-hiding menu entries**
+
+In `contributes.menus.commandPalette`, delete both objects (they exist only to hide the now-deleted declarations):
+
+```json
+        {
+          "command": "cowriting.acceptProposal",
+          "when": "false"
+        },
+        {
+          "command": "cowriting.rejectProposal",
+          "when": "false"
+        },
+```
+
+- [ ] **Step 3: Verify no dangling references**
+
+Run: `grep -rn "acceptProposal\|rejectProposal" package.json src/`
+Expected: no matches (the only remaining references are the retirement assertions in `test/e2e/suite-no-workspace/noWorkspace.test.ts`).
+
+Run: `node -e "JSON.parse(require('fs').readFileSync('package.json','utf8')); console.log('valid json')"`
+Expected: `valid json`
+
+- [ ] **Step 4: Commit**
+
+```bash
+git add package.json
+git commit -m "chore: delete orphaned acceptProposal/rejectProposal declarations (#71 item 2)"
+```
+
+---
+
+### Task 3: `openReviewPreview` editor/title when-key (item 3)
+
+**Files:**
+- Modify: `package.json:260` (the `editor/title` menu entry for `cowriting.openReviewPreview`)
+
+**Interfaces:** none — one-token when-clause fix. Title menus key off the resource, so `resourceLangId` is correct (every sibling `editor/title` entry already uses it); `editorLangId` is the odd one out.
+
+- [ ] **Step 1: Fix the key**
+
+In `contributes.menus."editor/title"`, change:
+
+```json
+        {
+          "command": "cowriting.openReviewPreview",
+          "when": "editorLangId == markdown",
+          "group": "navigation@9"
+        }
+```
+
+to:
+
+```json
+        {
+          "command": "cowriting.openReviewPreview",
+          "when": "resourceLangId == markdown",
+          "group": "navigation@9"
+        }
+```
+
+(Only the `editor/title` entry. The `commandPalette` entries elsewhere legitimately use `editorLangId`; the `editor/title/context` and `explorer/context` entries already use `resourceLangId`.)
+
+- [ ] **Step 2: Commit**
+
+```bash
+git add package.json
+git commit -m "fix: openReviewPreview title-menu when uses resourceLangId like its siblings (#71 item 3)"
+```
+
+---
+
+### Task 4: gate the `cowriting.createThread` palette entry (item 5)
+
+**Files:**
+- Modify: `package.json` (`contributes.menus.commandPalette`)
+
+**Interfaces:** none — `createThreadOnSelection` (`src/threadController.ts:197-201`) already returns `undefined` silently without a selection / coediting / authorable scheme; the palette entry is currently unconditionally visible, so the command silently no-ops from the palette. Fix = gate the palette entry with the **same when-clause its `editor/context` entry uses** (`package.json:290-293`), so the palette only offers it when it can act — consistent with how `markReviewed`'s palette gate was tightened at merge.
+
+- [ ] **Step 1: Add the palette gate**
+
+In `contributes.menus.commandPalette`, insert after the `cowriting.proposeAgentEdit` entry:
+
+```json
+        {
+          "command": "cowriting.createThread",
+          "when": "editorHasSelection && (resourceScheme == file || resourceScheme == untitled) && cowriting.isCoediting"
+        },
+```
+
+- [ ] **Step 2: Verify JSON**
+
+Run: `node -e "JSON.parse(require('fs').readFileSync('package.json','utf8')); console.log('valid json')"`
+Expected: `valid json`
+
+- [ ] **Step 3: Commit**
+
+```bash
+git add package.json
+git commit -m "fix: gate createThread palette entry on selection+coediting like its context-menu entry (#71 item 5)"
+```
+
+---
+
+### Task 5: clear the `gitBaseline` reflog debounce timer in `dispose()` (item 6)
+
+**Files:**
+- Modify: `src/gitBaseline.ts:113-126` (`watchReflog`)
+
+**Interfaces:** none — same disposable contract; the pushed disposable additionally clears the pending debounce timeout so no one-shot `repo.status()` fires post-dispose.
+
+- [ ] **Step 1: Clear the timer in the pushed disposable**
+
+In `watchReflog`, change:
+
+```ts
+      const watcher = fs.watch(reflog, () => {
+        if (pending) clearTimeout(pending);
+        pending = setTimeout(() => void repo.status().catch(() => {}), 150);
+      });
+      this.disposables.push({ dispose: () => watcher.close() });
+```
+
+to:
+
+```ts
+      const watcher = fs.watch(reflog, () => {
+        if (pending) clearTimeout(pending);
+        pending = setTimeout(() => void repo.status().catch(() => {}), 150);
+      });
+      this.disposables.push({
+        dispose: () => {
+          watcher.close();
+          if (pending) clearTimeout(pending);
+        },
+      });
+```
+
+- [ ] **Step 2: Verify**
+
+Run: `npm run typecheck`
+Expected: clean exit.
+
+- [ ] **Step 3: Commit**
+
+```bash
+git add src/gitBaseline.ts
+git commit -m "fix: clear reflog debounce timer on GitBaselineWatcher dispose (#71 item 6)"
+```
+
+---
+
+### Task 6: README F3 drift + stale chord notes in superseded banners (item 4)
+
+**Files:**
+- Modify: `README.md` (F3 section ~106, F7 banner ~230-233, F10 banner ~313-317)
+
+**Interfaces:** none — docs only. Current reality the banner must state: `Cowriting: Toggle Attribution` no longer exists (annotations toggle via **Toggle Annotations**, How it works §6); "Ask Claude to Edit Selection" is palette-hidden (`package.json:201-203` — `"when": "false"`), superseded by **Ask Claude to Edit** (`Ctrl+Alt+E` / `Cmd+Alt+E`) and the comments-first **Ask Claude** (§3). `Ctrl+Alt+R` today is **Review Changes** (native diff), not any preview.
+
+- [ ] **Step 1: Add the F3 banner**
+
+Immediately after the `## F3 — Live human/Claude attribution (Feature #6)` heading (before the "As you and Claude coauthor…" paragraph), insert:
+
+```markdown
+> **Commands superseded (native-surfaces migration).** The attribution **data
+> layer** below is current, but the Commands block is historical:
+> `Cowriting: Toggle Attribution` was retired — authorship/change coloring now
+> toggles via **Toggle Annotations** on a coedited document (How it works §6) —
+> and "Ask Claude to Edit Selection" is palette-hidden; use **Ask Claude to
+> Edit** (`Ctrl+Alt+E` / `Cmd+Alt+E`) or the comments-first **Ask Claude** (How
+> it works §3). Kept below as a historical record.
+```
+
+- [ ] **Step 2: Chord note in the F7 banner**
+
+Extend the F7 superseded banner's last sentence. Change:
+
+```markdown
+> **Superseded (Task 8, native-surfaces migration).** The bespoke webview this
+> section describes was deleted; its authorship/change coloring lives on
+> **inside VS Code's own built-in Markdown preview** — see **How it works** §6
+> above. Kept below as a historical record of the pre-migration design.
+```
+
+to:
+
+```markdown
+> **Superseded (Task 8, native-surfaces migration).** The bespoke webview this
+> section describes was deleted; its authorship/change coloring lives on
+> **inside VS Code's own built-in Markdown preview** — see **How it works** §6
+> above. Kept below as a historical record of the pre-migration design. (The
+> `Ctrl+Alt+R` chord below now opens the native **Review Changes** diff; the
+> annotated preview opens via **Open Cowriting Review Preview**, no chord.)
+```
+
+- [ ] **Step 3: Chord note in the F10 banner**
+
+Change the F10 banner's last sentence from:
+
+```markdown
+> **How it works** above. Kept below as a historical record.
+```
+
+to:
+
+```markdown
+> **How it works** above. Kept below as a historical record. (The `Ctrl+Alt+R`
+> chord below now opens the native **Review Changes** diff, not this panel.)
+```
+
+- [ ] **Step 4: Commit**
+
+```bash
+git add README.md
+git commit -m "docs: F3 superseded-commands banner + stale-chord notes in F7/F10 banners (#71 item 4)"
+```
+
+---
+
+### Task 7: full verification + PR + merge
+
+**Files:** none new.
+
+- [ ] **Step 1: Full local gate**
+
+Run each; all must be green:
+
+```bash
+npm test
+npm run typecheck
+npm run build
+npm run test:e2e
+```
+
+Expected: unit suite passes (~250 tests), typecheck clean, build succeeds, host E2E passes (undo suites may self-skip via the #54 runtime probe — a loud warning, not a failure).
+
+- [ ] **Step 2: Push branch + open PR**
+
+```bash
+git push -u origin worktree-s71-surface-polish
+```
+
+Open a PR against `main` on `benstull/vscode-cowriting-plugin` (Gitea, SSH remote) titled `Surface polish: batched Minor findings from the native-surfaces final review (#71)`; body lists the six items and cites the issue.
+
+- [ ] **Step 3: Merge (squash) and confirm #71 auto-close/close**
+
+Merge the PR; verify `main` carries the change and close #71 if the merge didn't.
@@ -0,0 +1,183 @@
+# Solution Design: Authorship view in the rendered preview (F9)
+
+| | |
+| --- | --- |
+| **Author(s)** | Ben Stull (with Claude) |
+| **Status** | `draft` |
+| **Version** | v0.1.0 |
+| **Anchor** | Feature **F9** (to be captured, ~#27) — builds on F7 `#21` (rendered preview) + F3 `#6` (live attribution). Surfaced as friction during F8 (`#25`) testing: "the preview doesn't show the Claude-composed annotations." |
+
+---
+
+## 1. Problem & context
+
+The F7 rendered track-changes preview (`src/trackChangesPreview.ts` +
+`src/trackChangesModel.ts`) answers **"what changed since the F6 baseline?"** It
+diffs `baselineText` vs the live buffer and renders author-*agnostic*
+`added`/`removed`/`changed` marks. It has never read F3 attribution.
+
+Two facts combine into the friction:
+
+1. The F6 baseline **advances on every machine landing** (INV-18): when a Claude
+   proposal is accepted, its text is folded *into* the baseline, so it then reads
+   as **unchanged** in track-changes.
+2. The preview has **no authorship axis** — it cannot say "Claude composed this
+   span" vs "you did."
+
+So a writer who accepts Claude's edits cannot *see* what Claude contributed in the
+preview. F3 attribution already records exactly that (Claude=`agent` vs human
+spans, char-precise), and F8 made attribution available on **any** authorable doc
+(in-folder / out-of-folder / untitled) — the data exists; the preview just doesn't
+show it.
+
+**Goal:** add an **Authorship** mode to the preview that renders the current
+document with each span colored by its F3 author — Claude (blue) vs you (green) —
+inline and char-precise, reading attribution directly (baseline-independent), so
+Claude's contributions are visible even after the baseline absorbs them.
+
+## 2. Solution overview
+
+The preview gains a **second mode**, switched by a **segmented control in the
+webview header** (`[ Track changes | Authorship ]`); mode is remembered per panel,
+default **Track changes** (today's behavior, unchanged).
+
+- **Track changes mode** — unchanged (`renderTrackChanges(baselineText, current)`).
+- **Authorship mode** — renders the **current buffer** (no baseline diff) via a new
+  pure engine `renderAuthorship(currentText, authorSpans)`, wrapping each
+  attributed span in `<span class="cw-by-claude">` / `<span class="cw-by-human">`.
+  A header **legend** (● Claude / ● You) appears in this mode. Text with no
+  attribution record renders plain.
+
+The two modes are distinct renderings of different axes; authorship mode does **not**
+combine with the baseline diff (no `ins`/`del`, no removed text — there is no diff).
+
+## 3. Technical design
+
+### 3.1 The render engine (pure, vscode-free — extends `trackChangesModel.ts`)
+
+New export `renderAuthorship(currentText: string, spans: AuthorSpan[]): string`,
+where `AuthorSpan = { start: number; end: number; author: "claude" | "human" }`
+(char offsets into `currentText`, non-overlapping). Deterministic (INV-22 extends
+to it): same inputs → identical HTML; no vscode, no DOM.
+
+Algorithm:
+
+1. **Block split with offsets.** Reuse `splitBlocks` but track each block's
+   `[start, end)` char range in `currentText` (add the offsets to the `Block`
+   shape, or a parallel `splitBlocksWithRanges`). Blank-line gaps between blocks
+   are outside any block range (consistent with today's whitespace handling).
+2. **Per block:**
+   - **Prose block:** clip `spans` to the block's range, translate to block-local
+     offsets, and inject paired **sentinel markers** at the local boundaries into
+     the block's raw markdown — four Unicode Private-Use Area code points: `U+E000`
+     (Claude-open) / `U+E001` (Claude-close), `U+E002` (human-open) / `U+E003`
+     (human-close). PUA code points never appear in real content and markdown-it
+     passes them through as plain text. Render the block with the existing
+     markdown-it instance, then **post-process** the HTML: string-replace
+     `U+E000`→`<span class="cw-by-claude">`, `U+E001`→`</span>`,
+     `U+E002`→`<span class="cw-by-human">`, `U+E003`→`</span>`.
+   - **Code / mermaid fence (atomic, INV-23 extends → INV-27):** never inject
+     sentinels inside a fence. If any span overlaps the fence, wrap the rendered
+     fence in a block div carrying the author class + a small badge
+     (`Claude` / `You` / `mixed` when both authors overlap); else render plain.
+3. Concatenate block HTML (same join as `renderTrackChanges`). Each block keeps
+   the existing per-block `try/catch` → error chip, so a render failure degrades
+   to a visible chip, never a throw.
+
+**Known edge case (documented):** a sentinel placed immediately adjacent to active
+emphasis markers (`**`, `_`) can perturb markdown-it's inline parsing for that
+span. Spans from F3 fall on real edit boundaries, so this is rare; v1 accepts it
+(the per-block chip prevents any hard failure) and covers the common cases with
+tests. Snapping span boundaries to token-safe positions is a deferred refinement.
+
+### 3.2 Data source & wiring
+
+- **Spans come from F3.** `AttributionController` already maintains live spans as
+  current-buffer char ranges with author kind (`getSpans(key)` → `RenderedSpan[]`,
+  `authorKind: "human" | "agent"`). Add `spansFor(document): AuthorSpan[]` to
+  `AttributionController` — it computes the document key internally
+  (`this.keyOf(document)` via the F8 router) and maps `agent→"claude"`,
+  `human→"human"`. This resolves the key mismatch (the preview keys panels by
+  `document.uri.toString()`; attribution keys by `keyOf`).
+- **Preview gains an attribution dependency.** `TrackChangesPreviewController`
+  takes `AttributionController` in its constructor. In `extension.ts`, construct
+  the authoring stack (router, guard, attribution) **before** the F7 controller
+  so the dependency is available (a small, safe reorder — F8 already removed the
+  no-root early return, so all controllers are constructed unconditionally).
+- **Mode state & toggle.** The controller holds `mode: Map<key, "changes"|"authorship">`
+  (default `"changes"`). `refresh(document)` renders the panel's current mode:
+  track-changes (existing path) or authorship (`renderAuthorship(current,
+  attribution.spansFor(document))`), posting `{type:"render", html, mode, epoch?,
+  summary?, legend?}`. The webview header shows the segmented toggle; clicking
+  posts `{type:"setMode", mode}` back; the controller updates the map and
+  re-renders. Edits/baseline-changes re-render in the current mode.
+- **Webview (`media/`):** add the segmented toggle to `#cw-header` and a
+  `setMode` message; render the posted HTML into `#cw-body` as today; show/hide
+  the epoch+summary (track-changes) vs the legend (authorship) per mode. CSS adds
+  `.cw-by-claude` (blue) / `.cw-by-human` (green) + the badge/legend styles,
+  themed via VS Code CSS variables (consistent with the existing preview CSS).
+
+### 3.3 Invariants (additions)
+
+- **INV-26 (authorship mode is baseline-independent):** authorship mode renders
+  the **current buffer** colored by F3 author, reading `AttributionController`,
+  never the F6 baseline. Track-changes mode is unchanged. The two are distinct
+  modes, never combined.
+- **INV-27 (fences stay atomic in authorship too):** a code/mermaid fence is never
+  marked inline; an overlapping author span yields a **block-level** author badge
+  (extends INV-23).
+- **INV-28 (pure authorship render):** `renderAuthorship(currentText, spans)` is
+  vscode-free and deterministic (extends INV-22) — spans are passed as data, so it
+  unit-tests with no editor.
+- INV-19/20/21 (read-only preview, sealed webview, no persistence/network) carry
+  over: authorship mode reads attribution, never mutates the document, sidecar, or
+  baseline; no new LLM/network/credential surface.
+
+## 4. Scope
+
+**In scope:** the Authorship mode + header toggle; `renderAuthorship` engine;
+`AttributionController.spansFor`; the preview↔attribution wiring + `extension.ts`
+reorder; webview toggle/legend/CSS; unit + host E2E + a manual-smoke addendum.
+
+**Out of scope / non-goals:** combining authorship with the diff in one view
+(rejected in design — two modes); marking removed text (no diff in authorship
+mode); the "untracked / always-there" third author class (rejected — only Claude &
+human are colored, unattributed renders plain); intra-emphasis sentinel-safety
+hardening (deferred); any change to F3 attribution capture, the seam, persistence,
+or the cross-rung contract.
+
+## 5. Testing strategy
+
+- **Unit (vitest, vscode-free):** `renderAuthorship` —
+  - single Claude span / single human span → correct wrapper class;
+  - two authors in one prose paragraph → exact inline boundaries;
+  - span clipped at a block boundary (spans one block only);
+  - a code fence and a mermaid fence overlapping a span → block-level badge,
+    no inner sentinels (atomic);
+  - adjacent same-author spans; a span covering a whole block;
+  - empty `spans` → plain render (equals markdown-it of the source);
+  - determinism (same inputs → identical HTML).
+- **Host E2E (`@vscode/test-electron`, no LLM, extends the F7 suite):** open the
+  preview on a markdown doc → land a Claude edit via the `proposeAgentEdit` seam +
+  accept → set authorship mode → assert the posted model marks Claude's span as
+  Claude-authored (and a human span as human). Assert track-changes mode still
+  renders as before (regression).
+- **Manual smoke (`docs/MANUAL-SMOKE-F9.md`):** the webview visuals — blue/green
+  colors, the header toggle, the legend, a mixed-author paragraph, a Claude-authored
+  mermaid fence badge — verified by eye (the sealed webview's rendering isn't
+  covered by the E2E).
+
+## 6. Delivery slices
+
+- **SLICE-1** `renderAuthorship` + block-offset split in `trackChangesModel.ts`
+  (the sentinel inject/post-process; atomic fences) + unit tests.
+- **SLICE-2** `AttributionController.spansFor` + wire it into
+  `TrackChangesPreviewController` (constructor dep, `extension.ts` reorder) +
+  per-panel mode state + the `setMode` message.
+- **SLICE-3** webview header toggle + legend + `.cw-by-*` CSS (`media/`).
+- **SLICE-4** host E2E (authorship-mode assertion + track-changes regression) +
+  `docs/MANUAL-SMOKE-F9.md` + README F9 note.
+
+**Done =** authorship mode shows Claude's spans (blue) and yours (green) inline on
+a markdown doc; the header toggle flips modes; code/mermaid fences get a block
+badge; track-changes mode unchanged; unit + host E2E green; smoke performed once.
@@ -0,0 +1,602 @@
+---
+status: graduated
+---
+# Solution Design: Preview Toolbar as the Primary Interaction Surface (F11)
+
+| | |
+| --- | --- |
+| **Author(s)** | Ben Stull (with Claude) |
+| **Reviewers / approvers** | Ben Stull |
+| **Status** | `draft` |
+| **Version** | v0.1.0 |
+| **Source artifacts** | Feature `benstull/vscode-cowriting-plugin#43` (F11, `type/feature`, `priority/P1`) · Epic `#1` (closed) · Capture session `vscode-cowriting-plugin-0035` (2026-06-12) · Brainstorming session `vscode-cowriting-plugin-0036` · Builds on (all shipped): F3 `#6` (live attribution), F4 `#12` (propose/accept), F6 `#17`/`#19` (baseline + diff view), F7 `#21`/`#22` (rendered preview), F9 `#27` (authorship preview), F10 `#29` (interactive review preview), F10-followups `#31` (inline-at-anchor proposals) · Coexists with `#41` (right-click → Open Review Panel) and `#42` (right-click → Ask Claude to Edit), both blocked-by this · Parent specs (graduated): `coauthoring-inner-loop.md`, `coauthoring-attribution.md`, `coauthoring-propose-accept.md`, `coauthoring-diff-view.md`, `coauthoring-rendered-preview.md`, `coauthoring-interactive-review.md` · Lineage: `ben.stull/rfc-app#48` |
+
+**Change log**
+
+| Date | Version | Change | By |
+| --- | --- | --- | --- |
+| 2026-06-12 | v0.1.0 | Initial draft — brainstorming session 0036 (from the capture in session 0035). Three forks locked with the operator: block-level preview-selection→source mapping; document edit diffed into per-hunk F4 proposals; #43 lands a minimal right-click→open-preview gateway. | Ben Stull + Claude |
+
+---
+
+## 1. Business Context
+
+### 1.1 Executive Summary
+
+F10 made the rendered preview the **single review surface** — clean editor on the
+left, annotated review on the right, with an annotations on/off toggle and
+inline ✓/✗ on Claude's pending proposals. But the writer still cannot *act* from
+the preview beyond accepting/rejecting proposals: to **ask Claude to edit**, they
+jump back to the editor and use a selection-gated context-menu item; to **pin a
+fresh review baseline** they have *no* reachable control at all (the
+`cowriting.pinDiffBaseline` command is registered but `when:false`, orphaned
+since `#34` removed its two-pane host); and whole-document editing doesn't exist.
+
+F11 makes the **preview toolbar the primary interaction surface**. Beside the
+existing annotations checkbox — the one control the writer already loves — the
+toolbar gains a **Pin baseline** button and a **single adaptive "Ask Claude…"
+button** that reads *Edit Selection* when text is selected in the preview and
+*Edit Document* when nothing is. The writer reads, asks Claude to edit, and
+resets the baseline all in one place, mouse-first, without leaving the rendered
+document. A minimal right-click entry opens the preview, making it the surface
+the `#41`/`#42` gateways will lead into.
+
+### 1.2 Background
+
+The inner loop shipped F2–F5 (threads · attribution · propose/accept ·
+cross-rung). F6 added the baseline + a native diff toggle; F7 the rendered
+track-changes preview; F9 an authorship mode; F10 (`#29`) collapsed those into the
+**single interactive review preview** (clean editor; annotations on/off; ✓/✗ on
+F4 proposals surfaced in the rendered view); `#31` then placed proposals
+**inline at their resolved anchor** in that preview.
+
+Capture session 0035 filed `#43` (this feature) plus `#41` (right-click → Open
+Review Panel) and `#42` (right-click → Ask Claude to Edit). The operator's ask:
+*"Can we set it up so all interactions — Ask Claude to Edit Selection / Edit
+Document, annotations off/on, Pin new baseline — are via the preview window? I
+like the annotations checkbox up there; make the others buttons, with one 'Ask
+Claude…' button that changes depending on whether some of the markdown preview is
+selected."* That session also surfaced the stranded `pinDiffBaseline` command
+(`when:false` since `#34`). This spec is the Solution Design for `#43`.
+
+### 1.3 Business Actors / Roles
+
+- **Coauthor (human)** — the markdown writer/engineer (PP-1); F11's sole user.
+- **Coauthor (machine)** — Claude via `@cline/sdk`; not a user of F11, but the
+  target of the toolbar's "Ask Claude…" gesture and the author of the proposals
+  that result.
+
+### 1.4 Problem Statement
+
+The plugin's interactions are scattered across surfaces and inconsistently
+reachable. "Ask Claude to Edit Selection" is only a selection-gated **editor**
+context-menu item; **whole-document editing doesn't exist**; and **Pin Review
+Baseline** is **unreachable from any UI**. The one control the writer loves — the
+annotations on/off checkbox in the preview — proves the toolbar is the natural
+home for these gestures, but it stands alone. A writer reviewing in the preview
+has to leave it and hunt through editor menus / the palette to act.
+
+### 1.5 Pain Points
+
+- **No edit gesture in the preview** — to ask Claude to change anything, the
+  writer leaves the review surface for the editor's right-click menu.
+- **Whole-document editing is missing** — there is no "edit the whole document"
+  path at all; only a selection-scoped editor command exists.
+- **Pin baseline is stranded** — the command exists but no menu, keybinding, or
+  palette entry reaches it (`when:false` since `#34`).
+- **Mouse-first review is broken mid-flow** — the preview is mouse-driven, but
+  acting forces a context-switch to keyboard/menus elsewhere.
+
+### 1.6 Targeted Business Outcomes
+
+The preview becomes a **self-contained cockpit** for the inner loop. From its
+toolbar a writer can toggle annotations (today), **ask Claude to edit** (selection
+or whole document, via one button that adapts to what's selected), and **pin a
+fresh baseline** — no context-switching to the editor or command palette. The
+interaction model consolidates around the surface the writer already prefers, and
+the stranded pin command gets a real home.
+
+### 1.7 Scope (business)
+
+**In scope:** preview-webview toolbar controls — a **Pin baseline** button and a
+**single adaptive "Ask Claude…" button** (Edit Selection ⇆ Edit Document) — beside
+the existing annotations checkbox; wiring those controls to the **existing** F4
+edit seam, F3 attribution, and F6 baseline command; **block-level**
+preview-selection → source-range mapping (the central design risk); a **new
+whole-document edit path** whose result is **diffed into per-hunk F4 proposals**;
+a **minimal right-click → Open Review Preview** entry so the surface is reachable
+end-to-end; resolving the pin-baseline reachability gap; unit + host-E2E coverage;
+manual webview smoke.
+
+**Out of scope (deferred, not forgotten):** **char-precise sub-block** selection
+mapping (block granularity is the locked v1 — §6.7); the **richer `#41`/`#42`
+menu sets** (this feature lands only the minimal gateway; `#41`/`#42` expand it);
+preview→source **scroll-sync** (`#32`); multi-file / batch editing; the Explorer
+tree affordance; any export / print / copy gesture.
+
+**Non-goals (firm):** **no new edit / attribution / proposal *model*** — F11
+reuses F3 `spansFor`, the F4 `propose`/`accept` single-range model, and the F6
+baseline store; no change to the **sidecar**, the **cross-rung contract**, or
+`SCHEMA_VERSION`; **no document mutation from the webview** (INV-20/21/34 hold —
+the sealed webview posts intent only); no LLM/network/credential surface added to
+the webview (INV-8 untouched — the edit turn runs host-side as today).
+
+### 1.8 Assumptions · Constraints · Dependencies
+
+- **Anchor:** Feature `#43` (F11). Builds directly on shipped work: the F7/F10
+  rendered preview + annotations toggle + host↔webview message bus, the F3
+  attribution + F4 propose/accept inner loop (including `#31`'s inline-at-anchor
+  proposal placement), and the F6 baseline store (`cowriting.pinDiffBaseline`,
+  currently unreachable — this feature gives it a home).
+- **Central design risk (locked):** the preview is a **rendered** sealed webview
+  (markdown-it HTML, strict CSP — F7 INV-21), so "Edit Selection" must map a
+  selection in the rendered preview back to a **source markdown range**. The
+  rendered HTML carries **no** source positions today; only internal block
+  char-offsets exist (`splitBlocksWithRanges` → `BlockWithRange.start/end`). The
+  locked approach is **block-level** mapping: the pure render layer emits
+  `data-src-start`/`data-src-end` on each rendered block; a selection resolves to
+  the union of the live-source blocks it intersects (§6.7, fork 1).
+- **Constraint (sealed webview):** interactive controls post messages to the
+  extension host, which applies edits/pins via the existing F4 / F6 paths; the
+  webview **never** edits the document, sidecar, or baseline directly. The edit
+  **turn** (LLM call) and the **instruction prompt** run host-side, keeping the
+  webview free of LLM/network/credential surface.
+- **Coexistence:** the native editor context-menu "Ask Claude to Edit Selection"
+  (`cowriting.editSelection`) stays unchanged; `#41`/`#42` add right-click
+  gateways *into* the preview. F11 lands only a minimal gateway (§6.7, fork 3).
+- No new persisted artifact; nothing in `.threads/`, the contract, or
+  `SCHEMA_VERSION` changes.
+
+### 1.9 Business Use Cases
+
+- **BUC-1 (edit from the preview)** Reviewing in the preview, the writer selects a
+  paragraph in the rendered document and clicks **"Ask Claude to Edit
+  Selection"**; Claude's proposed change appears as a blue ✓/✗ block at that
+  spot — without the writer ever leaving the preview.
+- **BUC-2 (edit the whole document)** With nothing selected, the writer clicks
+  **"Ask Claude to Edit Document"**, types an instruction, and Claude's rewrite
+  surfaces as **several** independently-acceptable blue proposal blocks (one per
+  changed hunk) inline in the preview.
+- **BUC-3 (pin a fresh baseline)** After accepting a batch of changes, the writer
+  clicks **"Pin baseline"**; the change-marks clear and "what changed" now counts
+  from this moment — the stranded command finally has a button.
+
+---
+
+## 2. Solution Proposal
+
+F11 is a **thin increment** on F10's preview: it adds two controls to the
+existing header toolbar and routes their intent through machinery that already
+exists. No new model, no new persisted state.
+
+**The pure render layer (`trackChangesModel.ts`) learns one new thing:** a shared
+helper wraps each rendered block in `<div data-src-start="N" data-src-end="M">`
+using the existing `splitBlocksWithRanges` offsets. It is applied to **both**
+render paths — `renderReview` (annotations on) and `renderPlain` (annotations
+off) — so selection→source mapping works in either mode. The helper is pure,
+vscode-free, DOM-free, and deterministic (extends INV-22/33). The render layer
+gains **no** selection or DOM logic.
+
+**The webview (`media/preview.ts` + `.css`)** header becomes:
+
+```
+[ ☑ Annotations ]   [ ⌖ Pin baseline ]   [ ✦ Ask Claude to Edit Document ▾ ]
+```
+
+The **Ask-Claude button morphs its own label** on `selectionchange`: a non-empty
+selection inside the rendered body → **"Ask Claude to Edit Selection"**; an empty
+/ collapsed selection → **"Ask Claude to Edit Document"**. On click it walks the
+selection's start and end nodes up to the nearest ancestor carrying
+`data-src-start`/`data-src-end` and posts the resolved offsets. That
+nearest-ancestor lookup is the webview's **only** mapping duty (manual-smoke
+territory); everything downstream is host-side and testable. The webview stays
+**sealed** (INV-21): nonce'd inline script, no network, no document mutation.
+
+**New webview→host messages (intent only):**
+
+```ts
+type ToolbarMsg =
+  | { type: "pinBaseline" }
+  | { type: "askClaude"; scope: "selection"; start: number; end: number }
+  | { type: "askClaude"; scope: "document" };
+```
+
+**The host (`trackChangesPreview.ts`)** routes each intent through the existing
+seams:
+
+- **`pinBaseline`** → pins the **previewed** document (calls
+  `DiffViewController.pin(document)` directly — not the `activeTextEditor`-based
+  command, which may not point at the previewed doc) → `onDidChangeBaseline` →
+  re-render with cleared marks.
+- **`askClaude`** → host `showInputBox` for the instruction (keeps the LLM /
+  secret surface out of the sealed webview), then one shared host routine
+  `runEditAndPropose(document, target, instruction)`:
+  - **selection** → `target` = the block-union range `[firstBlock.start …
+    lastBlock.end]`; one `runEditTurn` → one F4 `propose()` over that range (the
+    existing Edit-Selection shape).
+  - **document** → `target` = whole document; one `runEditTurn` over the full
+    text → **diff Claude's result against the current text → one `propose()` per
+    changed hunk** (multiple proposals, each its own blue ✓/✗ block). Reuses the
+    F4 single-range model N times; no model change.
+
+**The right-click gateway:** `cowriting.showTrackChangesPreview` is added to the
+`editor/title` menu (markdown only) so right-clicking the tab opens the preview —
+the minimal entry that makes the toolbar surface reachable end-to-end and lets
+`#41`/`#42` expand the menu set later.
+
+**Reachability cleanup:** `cowriting.pinDiffBaseline` gets a real palette `when`
+(`editorLangId == markdown`), resolving the orphan from the command side too; a
+new `cowriting.editDocument` command is registered (document-scoped edit) so
+`#42`'s gateway can reuse it.
+
+Everything downstream of *(intent) → (existing seam)* is the existing F4/F6/F3
+machinery; the only genuinely new pure code is the block-offset wrapper and the
+document-rewrite hunk-diff. Both are unit-testable with no vscode and no webview.
+
+---
+
+## 3. Product Personas
+
+- **PP-1 Inner-loop coauthor** — the human markdown writer/engineer (as F2–F10);
+  the only persona F11 serves.
+
+## 4. Product Use Cases
+
+- **PUC-1 (toolbar present)** Opening the review preview for a markdown document
+  shows the header with **three** controls: the annotations on/off checkbox
+  (existing), a **Pin baseline** button, and an adaptive **Ask Claude…** button.
+  Controls are inert (disabled) for non-authorable documents.
+- **PUC-2 (adaptive label)** With a non-empty selection in the rendered preview
+  body, the Ask-Claude button reads **"Ask Claude to Edit Selection"**; with no
+  selection it reads **"Ask Claude to Edit Document"**. The label flips live as
+  the selection changes.
+- **PUC-3 (edit selection)** The writer selects rendered text, clicks **Ask
+  Claude to Edit Selection**, and enters an instruction. The selection resolves to
+  the union of the source blocks it touches; Claude proposes a change over that
+  range; a single blue ✓/✗ proposal block appears inline at that anchor (`#31`).
+- **PUC-4 (edit document)** With nothing selected, the writer clicks **Ask Claude
+  to Edit Document**, enters an instruction; Claude rewrites the whole document;
+  the rewrite is diffed into hunks and surfaces as **N** independent blue ✓/✗
+  proposal blocks inline. Accepting/rejecting each is the F10 path unchanged.
+- **PUC-5 (pin baseline)** The writer clicks **Pin baseline**; the previewed
+  document's review baseline is pinned to now; the change-marks clear and the
+  `Since <epoch>` label updates. (No confirmation prompt — matches the existing
+  command's behavior; re-pinning is the recovery.)
+- **PUC-6 (right-click into the preview)** Right-clicking a markdown editor tab
+  shows **Open Review Preview**; choosing it opens the preview (the gateway
+  `#41`/`#42` will build upon).
+- **PUC-7 (graceful edges)** A selection confined to a deletion (struck) or
+  proposal block — which carries no live-source range — falls back to **document**
+  scope. An empty document or a selection that resolves to no live block → the
+  button stays in **Edit Document** mode. A non-authorable document → toolbar edit
+  controls are disabled. The LLM turn failing → the existing `runEditTurn`
+  error handling (no proposal created); the preview is unchanged.
+
+---
+
+## 5. UX Layout
+
+The F10 preview is unchanged except for its **header bar**, which now hosts three
+controls in a single row:
+
+- **☑ Annotations** — the existing on/off checkbox (kept first; the operator's
+  preferred control).
+- **⌖ Pin baseline** — a button; pins the previewed document's review baseline to
+  now and clears the change-marks.
+- **✦ Ask Claude to Edit Document ▾** — a single button whose label and behavior
+  adapt to the preview's selection state (Edit **Selection** when text is
+  selected, Edit **Document** otherwise). Clicking it opens a host input box for
+  the instruction.
+
+Buttons are styled with theme CSS variables (light / dark / high-contrast),
+matching the existing toolbar chrome; they sit in the same `#cw-toggle` header
+region as the annotations checkbox. When the previewed document is not authorable
+(F8 `isAuthorable`), the **Pin baseline** and **Ask Claude…** controls render
+**disabled** (the annotations toggle stays active — reading is always allowed).
+
+The rendered body is unchanged from F10/`#31`: green human additions, blue
+LLM-authored text, struck deletions, and pending Claude proposals as inline blue
+blocks with ✓/✗ at their resolved anchors. Proposals produced via the new toolbar
+edit gestures appear exactly as proposals do today.
+
+---
+
+## 6. Technical Design
+
+### 6.1 Invariants
+
+Parent invariants INV-1..INV-34 carry over unchanged. F11 adds:
+
+- **INV-35 (toolbar gestures route through existing seams; webview never
+  mutates)** The Pin baseline and Ask-Claude toolbar controls post **intent**
+  messages to the host; **all** mutation goes through the existing machinery — pin
+  via the F6 baseline store (`DiffViewController.pin`), edits via the F4
+  `propose` → `accept`/`applyAgentEdit` (`WorkspaceEdit`) seam with F3 attribution.
+  No divergent edit or baseline path is introduced. The sealed webview never
+  edits the document, sidecar, or baseline directly (INV-20/21/34 hold); the LLM
+  turn and the instruction prompt run host-side (INV-8 untouched).
+- **INV-36 (block-granular preview-selection → source mapping)** The pure render
+  layer emits `data-src-start`/`data-src-end` (source char offsets from
+  `BlockWithRange`) on **every** rendered block, in **both** the on (`renderReview`)
+  and off (`renderPlain`) modes. A preview selection resolves to the **union of
+  the live-source blocks it intersects** (`[min start … max end]`); blocks with no
+  live-source range (deletion-only / proposal blocks) are skipped, and a selection
+  that resolves to no live block falls back to **document** scope. The DOM
+  selection → nearest-`data-src` lookup is the webview's **sole** mapping duty;
+  the offsets and everything downstream (fingerprint, turn, propose) are host-side
+  and testable. The wrapping is deterministic — same inputs → identical HTML
+  (extends INV-22/33).
+- **INV-37 (single adaptive Ask-Claude button; scope-aware)** One toolbar button
+  serves both scopes. A non-empty live-source selection → **Edit Selection**: one
+  F4 proposal over the block-union range. An empty selection → **Edit Document**:
+  one `runEditTurn` over the whole document, its result **diffed into hunks**, one
+  F4 `propose()` per changed hunk. Both scopes call the same host
+  `runEditAndPropose` routine and reuse the F4 **single-range** proposal model
+  (the document case issues multiple single-range proposals — **no new model**).
+
+### 6.2 High-level architecture
+
+```mermaid
+flowchart LR
+  wv["webview header\n☑ Annotations · ⌖ Pin · ✦ Ask Claude (adaptive)"] -- "postMessage{pinBaseline | askClaude(scope,start?,end?)}" --> ctl["trackChangesPreview\n(vscode layer)"]
+  ctl -- "pin(document)" --> base["F6 DiffViewController\nbaseline store (INV-18)"]
+  ctl -- "showInputBox → runEditAndPropose" --> turn["runEditTurn\n(host-side LLM turn)"]
+  turn -- "selection: 1 replacement\ndocument: rewrite" --> ctl
+  ctl -- "selection → 1 propose()\ndocument → diff → N propose()" --> prop["F4 ProposalController\npropose() (single-range model)"]
+  prop -- "onDidChangeProposals" --> ctl
+  base -- "onDidChangeBaseline" --> ctl
+  ctl -- "(baseline, current, spans, proposals)" --> model["renderReview / renderPlain\n(pure)\n+ wrapBlocksWithSrc (NEW)"]
+  model -- "annotated HTML w/ data-src on blocks" --> ctl
+  ctl -- "postMessage{render}" --> wv
+```
+
+The dashed-in NEW pieces are: `wrapBlocksWithSrc` (pure), the
+`runEditAndPropose` host routine with its document-rewrite hunk-diff, the three
+inbound toolbar messages, and the `editor/title` gateway menu. Everything else is
+the existing F6/F4/F3/F10 machinery.
+
+### 6.3 Data model & ownership
+
+**No new persisted artifact** (INV-20). F11 adds only transient on-the-wire
+messages (the `ToolbarMsg` union in §2) and reuses F10's `RenderMsg`. Baseline is
+owned by F6, proposals by F4 (sidecar), attribution by F3 — all untouched. The
+block-offset `data-src` attributes are render-time only (not stored).
+
+### 6.4 Interfaces & contracts
+
+- **`trackChangesModel`** (vscode-free, pure): new
+  `wrapBlocksWithSrc(blocks: BlockWithRange[], renderedPerBlock: string[]):
+  string` (illustrative) — or, more precisely, both `renderReview` and
+  `renderPlain` route their per-block rendered HTML through a shared internal
+  helper that prepends `data-src-start`/`data-src-end` to each block's wrapping
+  element. Plus `diffToHunks(currentText: string, rewrittenText: string):
+  Array<{ start: number; end: number; replacement: string }>` — the pure
+  document-rewrite → per-hunk proposal-range list (vscode-free, deterministic).
+- **`TrackChangesPreviewController`** (vscode layer): handles the three new
+  inbound messages; gains a `runEditAndPropose(document, target: { kind:
+  "range"; start; end } | { kind: "document" }, instruction)` private routine;
+  takes (or reaches) the `DiffViewController` to pin the previewed doc and the
+  edit-turn entry. New test seams as needed (`getLastModel` already exists for
+  asserting marks without webview DOM).
+- **`DiffViewController`** (F6): `pin(document)` is reused as-is (the controller
+  already exposes pinning by document); `cowriting.pinDiffBaseline`'s
+  `package.json` `when` flips from `false` to `editorLangId == markdown`.
+- **`ProposalController`** (F4): `propose(...)` reused unchanged (called once for
+  selection, N times for a diffed document). No signature change.
+- **`AttributionController`** (F3): unchanged (`applyAgentEdit` reused on accept).
+- **Commands / menus (`package.json`):**
+  - `cowriting.showTrackChangesPreview` — added to `editor/title` with
+    `when: editorLangId == markdown` (the minimal gateway). Existing palette +
+    `ctrl+alt+r` kept.
+  - `cowriting.pinDiffBaseline` — `when` flips to `editorLangId == markdown`
+    (no longer orphaned).
+  - `cowriting.editDocument` ("Ask Claude to Edit Document", document-scoped) —
+    new command registered, routed through `runEditAndPropose({kind:"document"})`;
+    available for `#42` to reuse. (The preview's selection-scoped edit is driven
+    by the `askClaude` message carrying webview-resolved offsets, not a command,
+    since the offsets originate in the webview.)
+- **Webview asset** (`media/preview.ts` + `.css`): header gains the two buttons;
+  a `selectionchange` listener updates the Ask-Claude label; click handlers post
+  the `ToolbarMsg` intents; the selection→nearest-`data-src` lookup helper. Stays
+  sealed (nonce'd inline script, CSP unchanged).
+
+### 6.5 Per–Product-Use-Case design
+
+- **PUC-1 (toolbar present):** render the two buttons in the header next to the
+  annotations checkbox; disable Pin + Ask-Claude when `!isAuthorable(document)`.
+- **PUC-2 (adaptive label):** webview `selectionchange` → if the selection is
+  non-empty and within the rendered body, label = "Edit Selection"; else "Edit
+  Document". Pure webview-local state.
+- **PUC-3 (edit selection):** webview resolves selection → `{start, end}` from the
+  nearest `data-src` ancestors → `postMessage{askClaude, selection, start, end}` →
+  host `showInputBox` → `runEditAndPropose({kind:"range", start, end})` →
+  `runEditTurn` → one `propose()` → `onDidChangeProposals` → re-render (inline
+  blue block at the anchor, `#31`).
+- **PUC-4 (edit document):** `postMessage{askClaude, document}` → host input box →
+  `runEditAndPropose({kind:"document"})` → `runEditTurn` over full text →
+  `diffToHunks(current, rewritten)` → one `propose()` per hunk → re-render (N blue
+  blocks).
+- **PUC-5 (pin baseline):** `postMessage{pinBaseline}` → `DiffViewController.pin(
+  previewedDocument)` → `onDidChangeBaseline` → re-render (marks cleared).
+- **PUC-6 (right-click gateway):** `editor/title` entry invokes
+  `cowriting.showTrackChangesPreview` for the tab's document.
+- **PUC-7 (edges):** selection resolving to no live block → document scope;
+  non-authorable → controls disabled; `runEditTurn` failure → existing error path,
+  no proposal; empty doc → Edit Document over empty range (no-op-safe).
+
+### 6.6 Non-functional requirements & cross-cutting concerns
+
+The webview stays **sealed** (INV-21): local assets, strict CSP with a per-load
+nonce, no network; the new inline handlers only read `data-src`/`data-proposal-id`
+and post intent (no eval, no remote, no document mutation). The instruction prompt
+and the LLM turn remain **host-side** — the webview gains **no** LLM, network, or
+credential surface (INV-8 untouched). `diffToHunks` and the block wrapping are
+O(document), run on a host gesture (not per-keystroke), fine at inner-loop scale.
+No telemetry, nothing persisted.
+
+### 6.7 Key decisions & alternatives considered
+
+| Decision | Chosen | Alternatives rejected |
+| --- | --- | --- |
+| **Preview-selection → source mapping granularity** | **Block-level** — pure layer emits `data-src-start/end` from existing `BlockWithRange`; selection → union of intersected live-source blocks. Robust, reuses what exists, ships the full adaptive button now. *(Operator decision, session 0036.)* | **Char-precise** sub-block mapping — needs per-inline-token source offsets markdown-it doesn't reliably give; rendered text ≠ source (syntax stripped) → fragile, risks the whole feature on the hardest part. **Document-only first** — defers the headline adaptive button; punts the risk. |
+| **Document-edit proposal granularity** | **Diff Claude's rewrite into hunks → one F4 proposal per changed hunk** — independent ✓/✗ per change; reuses the single-range model N times (no model change). *(Operator decision, session 0036.)* | **One whole-document proposal** — a single giant blue block, all-or-nothing accept/reject; poor UX for a real rewrite. |
+| **`#43` vs `#41`/`#42` scope** | **`#43` lands a minimal right-click → Open Review Preview gateway** (`editor/title`), so the toolbar surface is reachable end-to-end and its E2E is real; `#41`/`#42` expand the menu set. *(Operator decision, session 0036.)* | **Toolbar only; all menus in `#41`/`#42`** — `#43`'s "a right-click entry opens the preview" acceptance/E2E couldn't be satisfied within `#43`. |
+| **Instruction prompt location** | **Host `showInputBox`** — keeps LLM/secret surface out of the sealed webview; reuses the existing edit-turn flow. | **In-webview text field** — pushes prompt handling toward the sandbox; no benefit. |
+| **Pin button target** | **The previewed document** (`DiffViewController.pin(document)`) — the preview knows its bound doc. | **`activeTextEditor`-based command** — may not point at the previewed doc; the source of the orphan. |
+| **Pin confirmation** | **No confirm** — matches the existing command; re-pinning recovers. | **Confirm dialog** — friction for a routine, recoverable gesture. |
+
+### 6.8 Testing strategy
+
+- **Unit (vitest, vscode-free):** `data-src-start/end` present and correct on every
+  block for **both** `renderReview` and `renderPlain` (offsets equal the
+  `BlockWithRange` ranges; determinism — same inputs → identical HTML);
+  `diffToHunks` over fixtures — a single-hunk rewrite → one range; a multi-hunk
+  rewrite → the expected disjoint ranges with correct replacements; an unchanged
+  rewrite → zero hunks; whole-document replacement → one full-range hunk.
+- **Host E2E (`@vscode/test-electron`, no LLM, extends the F10 suite):** open a
+  markdown fixture → `cowriting.showTrackChangesPreview`. Simulate
+  `{type:"pinBaseline"}` → `getLastModel` shows cleared change-marks + advanced
+  epoch. Simulate `{type:"askClaude", scope:"selection", start, end}` with a
+  stubbed edit turn → exactly **one** proposal over the resolved range, anchored
+  inline. Simulate `{type:"askClaude", scope:"document"}` with a stubbed
+  multi-hunk rewrite → **N** proposals matching the hunks. Invoke the
+  `editor/title` gateway command → panel opens. Non-authorable document → toolbar
+  edit controls disabled (asserted via the model/flags the host exposes). The
+  webview DOM, real button clicks, the `selectionchange` label flip, and the
+  selection→`data-src` lookup are **not** E2E-asserted (sealed sandbox) — manual
+  smoke.
+- **Live smoke (manual — `docs/MANUAL-SMOKE-F11.md`):** open a markdown doc; open
+  the review preview; confirm the three header controls; select a paragraph →
+  button reads "Edit Selection", click → enter instruction → a blue ✓/✗ block
+  appears at that paragraph; clear the selection → button reads "Edit Document",
+  click → instruction → several blue blocks appear; click **Pin baseline** → marks
+  clear, `Since` label updates; right-click the tab → **Open Review Preview** opens
+  the panel; verify light/dark theming and that `git status` shows nothing
+  unexpected.
+
+### 6.9 Failure modes, rollback & flags
+
+A selection that resolves to no live-source block → **Edit Document** scope (never
+an error). `runEditTurn` failing → existing error handling, no proposal created,
+preview unchanged. `diffToHunks` producing zero hunks (rewrite == current) → no
+proposals, a brief "no changes proposed" host notice. Webview disposed mid-gesture
+→ the host routine completes against the document; the next open re-renders.
+**No feature flag** — the toolbar controls are additive UI; nothing persists.
+Rollback is reverting the PR with **zero** data migration (nothing persisted; the
+F6 baseline, F4 sidecar, F3 attribution data are untouched; the unhidden pin
+command and the gateway menu simply disappear).
+
+---
+
+## 7. Delivery Plan
+
+### 7.1 Approach / strategy
+
+One planning-and-executing session (F11 = `#43`), plan written just-in-time from
+this spec — the F2–F10 precedent. Host-E2E tier (a VS Code extension has no
+browser/deploy stage); no LLM in CI (edit turns stubbed). The webview's visual
+rendering, the adaptive label, and the selection→source DOM lookup are verified by
+the manual smoke; the automated seams are the pure block-wrapping + `diffToHunks`
+model and the host's message→seam wiring.
+
+### 7.2 Slicing plan
+
+- **SLICE-1 — Pin baseline button + reachability.** Webview header **Pin
+  baseline** button → `{type:"pinBaseline"}` → host `DiffViewController.pin(
+  previewedDoc)`; unhide `cowriting.pinDiffBaseline` (`when: editorLangId ==
+  markdown`). Host E2E: pin message clears marks. *(Immediate win — homes the
+  orphaned command.)*
+- **SLICE-2 — Block-offset emission.** Shared pure helper wrapping each block with
+  `data-src-start/end` in `renderReview` **and** `renderPlain`; vitest for both
+  modes + determinism. No UI yet. (INV-36 data layer.)
+- **SLICE-3 — Edit Document button + hunk path.** Webview **Ask Claude to Edit
+  Document** button (no-selection state) → `{type:"askClaude", scope:"document"}`;
+  host `runEditAndPropose({document})` → `runEditTurn` → `diffToHunks` → N
+  `propose()`; register `cowriting.editDocument`; vitest for `diffToHunks`; host
+  E2E for the N-proposal path. (INV-37 document half.)
+- **SLICE-4 — Adaptive Edit Selection.** Webview `selectionchange` label flip +
+  selection→nearest-`data-src` resolution → `{type:"askClaude", scope:"selection",
+  start, end}`; host single-range `propose()`. Host E2E for the selection message →
+  one anchored proposal. (INV-37 selection half; INV-36 consumer.)
+- **SLICE-5 — Gateway, edges, tests & docs.** `editor/title` → Open Review Preview
+  gateway; non-authorable disabling; host E2E (gateway opens panel; controls
+  inert on non-authorable); `docs/MANUAL-SMOKE-F11.md`; README F11 section.
+
+E2E are first-class plan tasks (handbook §9/§4); this app's required tier is host
+E2E (the F2–F10 precedent).
+
+### 7.3 Rollout / launch plan
+
+Non-shippable (no marketplace publish). "Done" = `#43` acceptance: the preview
+toolbar hosts the annotations checkbox + a Pin baseline button + a single adaptive
+Ask-Claude button (Edit Selection ⇆ Edit Document) that route through the existing
+F4/F3/F6 machinery; edits surface as proposals (one for a selection, per-hunk for
+a document rewrite); a right-click entry opens the preview; the pin command is no
+longer orphaned; unit + host E2E green; live smoke performed once.
+
+### 7.4 Risks & mitigations
+
+| Risk | Mitigation |
+| --- | --- |
+| Block-level selection feels coarse vs the editor's char-precise Edit Selection | Locked v1 decision (§6.7); a rendered surface is naturally block-grained; char-precise is a deferred follow-up if the coarseness bites |
+| `data-src` attributes perturb markdown-it output or the F10 proposal/diff rendering | Wrapping is applied at the block boundary (outside inline parsing); covered by determinism + both-mode unit tests; per-block `try/catch` error chip (F7) on render failure |
+| `diffToHunks` produces awkward hunk boundaries on a large rewrite | Pure + unit-tested over fixtures; hunks are line/block-aligned; worst case is more/fewer blocks, all independently ✓/✗-able — never wrong, just granular |
+| Selection inside a deletion/proposal block has no live-source range | Falls back to Document scope by design (INV-36); manual-smoke verified |
+| The webview selection→`data-src` lookup isn't E2E-testable (sealed) | The host half (offsets→fingerprint→propose) is E2E'd via simulated messages; the DOM lookup is the only manual-smoke-only seam, kept deliberately thin |
+| Unhiding pin / adding `editDocument` widens the command surface | Both guard on `editorLangId == markdown`; both route through existing seams; no new model or persisted state |
+
+---
+
+## 8. Traceability matrix
+
+| Requirement (`#43`) | Use case | Design | Slice |
+| --- | --- | --- | --- |
+| Pin baseline button in the preview toolbar | PUC-5 | INV-35, §6.4 (`DiffViewController.pin`) | SLICE-1 |
+| Resolve the orphaned `pinDiffBaseline` reachability | PUC-5 | §6.4 (`when` flip) | SLICE-1 |
+| Single adaptive Ask-Claude button (Selection ⇆ Document) | PUC-2/3/4 | INV-37, §6.2 | SLICE-3/4 |
+| Preview-selection → source range mapping (block-level) | PUC-3 | INV-36, §6.7 | SLICE-2/4 |
+| Edit Document path (new whole-document edit) | PUC-4 | INV-37, §6.5 (hunk diff) | SLICE-3 |
+| Edits route through existing F4/F3 (no divergent path) | PUC-3/4 | INV-35, §6.4 | SLICE-3/4 |
+| Right-click entry opens the preview (minimal gateway) | PUC-6 | §6.4 (`editor/title`) | SLICE-5 |
+| Controls only active for supported (authorable) docs | PUC-1/7 | §6.5 | SLICE-5 |
+| Sealed webview, no document mutation / LLM surface | — | INV-21/35, §6.6 | all |
+| No new edit/attribution/proposal model | — | §1.7, INV-37 | all |
+| Unit + host E2E + right-click-opens-preview coverage | — | §6.8 | SLICE-1..5 |
+
+## 9. Open Questions & Decisions log
+
+- **RESOLVED (session 0036, operator):** preview-selection → source mapping =
+  **block-level** (`data-src` attributes from `BlockWithRange`; union of
+  intersected blocks); document edit = **diffed into per-hunk F4 proposals**;
+  `#43` **lands a minimal right-click → Open Review Preview gateway** (`#41`/`#42`
+  expand the menus).
+- **RESOLVED (this spec, autonomous):** instruction prompt = **host `showInputBox`**
+  (LLM/secrets stay out of the webview); Pin targets the **previewed document**
+  (`DiffViewController.pin`); **no confirmation** on pin (matches existing); a new
+  `cowriting.editDocument` command is registered for `#42` reuse; `pinDiffBaseline`
+  is unhidden (`editorLangId == markdown`).
+- **OPEN → later:** **char-precise** sub-block selection mapping (deferred — block
+  granularity is v1); the **richer `#41`/`#42` menu sets** (this lands only the
+  minimal gateway); preview→source **scroll-sync** (`#32`); whether a large
+  document rewrite should cap/segment its hunks (only if real rewrites prove
+  noisy); the repo rename to `vscode-markdown-cowriting-plugin` (`#35`, deferred).
+
+## 10. Glossary & References
+
+- **Preview toolbar** — the review preview's header row: the annotations on/off
+  checkbox (existing) plus F11's Pin baseline and adaptive Ask-Claude buttons.
+  **Adaptive Ask-Claude button** — one button reading "Edit Selection" (non-empty
+  preview selection) or "Edit Document" (none). **Block-level selection mapping** —
+  resolving a rendered-preview selection to the union of source blocks it
+  intersects, via `data-src-start/end` attributes emitted by the pure render
+  layer. **Hunk-diffed document edit** — Claude's whole-document rewrite split
+  into changed hunks, each surfaced as its own F4 proposal. **Pin baseline** — F6's
+  `DiffViewController.pin` applied to the previewed document. **Gateway** — a
+  right-click entry that opens the preview (this feature lands the minimal one;
+  `#41`/`#42` expand them).
+- Feature `#43` (F11) · Epic `#1` · builds on F3 `#6`, F4 `#12`, F6 `#17`/`#19`,
+  F7 `#21`/`#22`, F9 `#27`, F10 `#29`/`#31` · coexists with `#41`/`#42` · parent
+  specs `coauthoring-inner-loop.md`, `coauthoring-attribution.md`,
+  `coauthoring-propose-accept.md`, `coauthoring-diff-view.md`,
+  `coauthoring-rendered-preview.md`, `coauthoring-interactive-review.md` · capture
+  session 0035 · lineage `ben.stull/rfc-app#48`.
@@ -0,0 +1,150 @@
+# Author-colored track changes (both panes) — design
+
+**Date:** 2026-06-26
+**Status:** Design approved; ready for implementation planning
+**Mockups:** `.superpowers/brainstorm/15095-1782504559/content/panes-states-v2.html`
+
+## Problem
+
+Today the plugin uses two unrelated color systems:
+
+- **The diff** (proposed/changed text in both panes) uses *semantic* colors —
+  green = added, red = removed — regardless of who made the change.
+- **Authorship** (F3) uses *author* colors (blue = Claude, green = human) but
+  only as a flat tint in the **preview** pane, and only on committed text, not
+  as a diff.
+
+So you cannot look at a change and immediately see **who** made it. The product
+goal is an experience like two humans collaborating in a track-changes document
+(Google Docs suggesting mode / Word track changes): every edit is a diff, and a
+glance tells you both *what changed* and *who changed it*.
+
+## The model
+
+Every change is rendered as a diff — insertions **and** deletions — where:
+
+- **Style encodes the operation:** underline = inserted, strikethrough = removed.
+- **Color encodes the author.**
+
+|        | Added            | Removed                |
+| ------ | ---------------- | ---------------------- |
+| Human  | green underline  | red strikethrough      |
+| Claude | blue underline   | purple strikethrough   |
+
+Human keeps today's exact green/red. Claude gets blue/purple (blue is already
+Claude's established attribution hue; purple reads as a removal without colliding
+with human red).
+
+**Color means "changed since the pinned baseline,"** not "authored forever." Once
+a change is accepted and the baseline re-pins, its marks clear and the text
+returns to plain. This preserves the existing pin→clean behavior (#48): a pinned,
+zero-diff document renders with no annotation.
+
+### Concrete colors
+
+Carry the existing human values; add Claude's pair. (Final hex values are a
+polish detail for implementation; these are the design intent.)
+
+- Human added: `#3fb950` underline, bg `rgba(63,185,80,0.14)`
+- Human removed: `#f85149` strikethrough, bg `rgba(248,81,73,0.11)`
+- Claude added: `#58a6ff` underline, bg `rgba(88,166,255,0.15)`
+- Claude removed: `#bc8cff` strikethrough, bg `rgba(188,140,255,0.13)`
+
+Each change run is a subtle author-hued background tint **plus** a solid 2px
+underline (insert) or strikethrough (delete) in the full author color — the solid
+line carries the signal even where backgrounds are faint.
+
+## Both panes (decision A)
+
+Author-colored track changes appear in **both** the main editor pane and the
+rendered preview pane.
+
+- **Main editor pane:** gains full inline track-changes for *all*
+  changes-since-baseline — human and Claude, committed and pending — not just
+  pending proposals. **This reverses INV-32** ("editor fully clean") from F10:
+  the editor was deliberately kept clean, with all who-wrote-what coloring living
+  only in the preview. We are intentionally undoing that.
+- **Rendered preview pane:** shows the same model (it already renders a diff +
+  authorship; this unifies them so ins/del runs are author-colored rather than
+  semantically colored).
+
+Pending Claude proposals keep their Accept/Reject affordance in both panes —
+CodeLens (`✓ Accept ▾ / ✗ Reject ▾`) in the editor, buttons in the preview.
+
+## States (must all render correctly)
+
+Base states: human-added, human-removed, Claude-added, Claude-removed.
+
+**Overlap** — a single span both authors touched — stacks both marks:
+
+- Human added it, Claude is removing it → green underline **+** purple strikethrough.
+- Claude added it, human is removing it → blue underline **+** red strikethrough.
+
+Block-level: a whole added block carries the author's add treatment (left border +
+tint); a whole removed block carries the author's remove treatment (struck + tint).
+
+Resting state: unchanged-since-baseline text renders plain. After Accept + re-pin,
+changed text returns to plain.
+
+## Architecture
+
+The core change is **fusing the currently-separate semantic-diff and authorship
+systems** so that every inserted/deleted run carries its author, and rendering
+colors by that author rather than by add/remove semantics.
+
+Components affected (per the current code map):
+
+- **`src/trackChangesModel.ts`** — `renderReview` / `renderOp` /
+  `wordMergedMarkdown` / `colorByAuthor`. Today `<ins>`/`<del>` are emitted with
+  semantic green/red classes independent of author. Change: ins/del runs become
+  author-aware and emit author-colored classes; overlap emits a stacked-mark
+  class.
+- **`src/editorProposalController.ts`** — today only pending proposals decorate,
+  using semantic `ThemeColor`s. Change: per-author/op `TextEditorDecorationType`s
+  (human-add/human-del/claude-add/claude-del + overlap), and decorate **all**
+  changes-since-baseline, not just proposals (the INV-32 reversal).
+- **`src/attributionController.ts`** — supplies the per-character/per-run author
+  spans the diff needs to attribute insertions (and, see below, deletions).
+- **`media/preview.css`** — recolor: keep human green/red, add Claude
+  blue/purple, encode underline=insert / strikethrough=delete, add overlap
+  (stacked-mark) classes for both directions.
+
+## Open technical question (for the plan / tech-design stage)
+
+**Deletion authorship.** Removed text is no longer in the buffer, so the renderer
+must know *who deleted it* to color the strikethrough.
+
+- For **pending proposals**, the deleter is trivially Claude (the proposal owns
+  the deletion).
+- For **committed deletions** (baseline → current), we must confirm the F3 seam /
+  attribution model carries the **deleting actor**. F3 attributes
+  *current-buffer* characters; deleted characters are gone. If the seam's edit
+  events carry the actor that performed each removal, we attribute from that; if
+  not, deriving deletion authorship is the main thing the implementation plan has
+  to solve. Until resolved, a safe fallback is to color committed deletions by the
+  **turn actor** that produced the diff hunk.
+
+This is the one genuinely open architectural item; everything else is
+presentation.
+
+## Out of scope (YAGNI)
+
+- More than two authors / arbitrary per-collaborator palettes. Two roles
+  (human, Claude) only.
+- User-configurable colors / theme settings. Ship one good scheme.
+- Changing accept/reject *mechanics* — this is purely how changes are
+  **visualized**; the F4 proposal lifecycle is unchanged.
+- Per-word authorship inside an accepted, re-pinned (plain) document.
+
+## Testing
+
+- **Unit (pure render):** `trackChangesModel` renders each state to the expected
+  author-colored markup — the four base states, both overlap directions, block
+  add/remove, and resting/plain. Reuse the existing pure-render test pattern.
+- **Editor decorations:** the decoration plan produces the right per-author/op
+  ranges, including the INV-32-reversal (committed changes now decorate) and
+  stacked overlap ranges.
+- **Host E2E:** open a co-edited doc, assert both panes show author-colored
+  marks; accept a proposal + re-pin → marks clear (pin→clean preserved).
+- Update/replace any test that asserts the **old** semantic green=add/red=remove
+  classes or the INV-32 clean-editor behavior.
@@ -30,13 +30,33 @@ const liveTurnOptions = {
  logLevel: "info",
 };

+/** @type {import('esbuild').BuildOptions} */
+const previewMermaidOptions = {
+  entryPoints: ["media/preview-mermaid.ts"],
+  outfile: "out/media/preview-mermaid.js",
+  bundle: true,
+  // Task 7 (Q4): contributed via markdown.previewScripts into the BUILT-IN
+  // preview's webview (a browser context) — mermaid is bundled IN (and ONLY
+  // in) this asset so it never bloats the extension-host bundle (the
+  // @cline/sdk size discipline).
+  platform: "browser",
+  format: "iife",
+  target: "es2020",
+  sourcemap: true,
+  logLevel: "info",
+};
+
 if (watch) {
  const ctx = await context(options);
  const ctxLive = await context(liveTurnOptions);
-  await Promise.all([ctx.watch(), ctxLive.watch()]);
+  const ctxPreviewMermaid = await context(previewMermaidOptions);
+  await Promise.all([ctx.watch(), ctxLive.watch(), ctxPreviewMermaid.watch()]);
  console.log("esbuild: watching…");
 } else {
  await build(options);
  await build(liveTurnOptions);
-  console.log("esbuild: build complete → out/extension.cjs + out/liveTurn.mjs");
+  await build(previewMermaidOptions);
+  console.log(
+    "esbuild: build complete → out/extension.cjs + out/liveTurn.mjs + out/media/preview-mermaid.js",
+  );
 }
@@ -0,0 +1,55 @@
+/*
+ * Task 7 (D3/D21, PUC-3) — authorship + change annotations for the BUILT-IN VS
+ * Code Markdown preview (`markdown.previewStyles`). Same F10 vocabulary as the
+ * (sunsetting) custom webview's `media/preview.css`: style = operation
+ * (underline = inserted, strikethrough = removed), color = author (human
+ * green, Claude blue) — `cw-del` has no author variant (a single fixed
+ * struck-red mark; see `previewAnnotations.ts`/`trackChangesModel.ts`'s "del"
+ * sentinel tag). Light-theme defaults below; `body.vscode-dark` (also applied
+ * on high-contrast dark) overrides with the same palette the webview preview
+ * already ships, for continuity across both review surfaces.
+ */
+.cw-ins-claude {
+  background: rgba(9, 105, 218, 0.12);
+  border-bottom: 2px solid #0969da;
+  text-decoration: none;
+}
+.cw-ins-human {
+  background: rgba(26, 127, 55, 0.12);
+  border-bottom: 2px solid #1a7f37;
+  text-decoration: none;
+}
+.cw-del {
+  background: rgba(207, 34, 46, 0.1);
+  text-decoration: line-through;
+  text-decoration-color: #cf222e;
+  opacity: 0.75;
+}
+
+body.vscode-dark .cw-ins-claude,
+body.vscode-high-contrast:not(.vscode-high-contrast-light) .cw-ins-claude {
+  background: rgba(88, 166, 255, 0.15);
+  border-bottom-color: #58a6ff;
+}
+body.vscode-dark .cw-ins-human,
+body.vscode-high-contrast:not(.vscode-high-contrast-light) .cw-ins-human {
+  background: rgba(63, 185, 80, 0.14);
+  border-bottom-color: #3fb950;
+}
+body.vscode-dark .cw-del,
+body.vscode-high-contrast:not(.vscode-high-contrast-light) .cw-del {
+  background: rgba(248, 81, 73, 0.11);
+  text-decoration-color: #f85149;
+}
+
+/*
+ * F7.1 (#22) intra-diagram mermaid diff legend, shown beneath a mermaid fence
+ * `previewAnnotations.ts` re-emitted through `mermaidDiff.ts` (Task 7 §2.6
+ * parity — same legend markup + colors as the sunset webview's
+ * `media/preview.css`).
+ */
+.cw-mermaid-legend { display: flex; gap: 0.6rem; font-size: 0.75em; opacity: 0.85; margin: 0.2rem 0 0.6rem; }
+.cw-mermaid-legend .cw-leg { padding: 0 0.4em; border-radius: 3px; border: 1px solid; }
+.cw-mermaid-legend .cw-leg-add { color: #2ea043; border-color: #2ea043; }
+.cw-mermaid-legend .cw-leg-chg { color: #d29922; border-color: #d29922; }
+.cw-mermaid-legend .cw-leg-rem { color: #808080; border-color: #808080; border-style: dashed; }
@@ -0,0 +1,57 @@
+/**
+ * Task 7 (Q4 mermaid check, D3/D21): the built-in VS Code Markdown preview does
+ * not render mermaid fences on its own — `previewAnnotations.ts`'s
+ * `cowritingMarkdownItPlugin` teaches ```mermaid fences to render as
+ * `<pre class="mermaid">SRC</pre>` (via `options.highlight`, the same
+ * extension point the built-in preview's own fence rule already calls); this
+ * script (contributed via `markdown.previewScripts`) is what actually turns
+ * those into diagrams — mermaid needs a DOM, so it runs here, in the preview's
+ * webview, never in the extension host. Bundled by esbuild as a standalone
+ * IIFE → out/media/preview-mermaid.js, so mermaid never enters the host
+ * bundle (matching the sealed webview's `media/preview.ts` precedent).
+ *
+ * Contributed preview scripts are reloaded on every content change (VS Code
+ * docs), so running at module top-level is sufficient; the
+ * `vscode.markdown.updateContent` listener is added defensively to also cover
+ * in-place content updates that don't reload the script (mirrors the proven
+ * `bierner.markdown-mermaid` extension's own approach).
+ *
+ * RESOLVED SCOPE NOTE (Q4 finding, migration plan Task 7 Step 2.6; wired in a
+ * cross-task review follow-up): this script itself only turns `pre.mermaid`
+ * nodes into diagrams — it never sees a diff. The intra-diagram diff/legend
+ * augmentation (F7.1, INV-29..31) happens one step earlier, source-side,
+ * where it CAN be unit-tested without a DOM: `previewAnnotations.ts`'s
+ * `cowritingMarkdownItPlugin` re-emits a CHANGED mermaid fence's source
+ * through `mermaidDiff.ts` (`buildMermaidQueue`, block-level `diffBlocks`
+ * pairing against the F6/F7 baseline) before this script ever runs, so the
+ * `pre.mermaid` node this script hands to `mermaid.run` already carries the
+ * `classDef`/`class`/`linkStyle` styling directives — this script needn't (and
+ * can't, DOM-less-ly) know a diagram changed at all. Basic mermaid rendering
+ * (this file) is exercised by the same proven pattern as the real
+ * `bierner.markdown-mermaid` extension; the augmentation upstream is exercised
+ * by pure unit tests (`test/previewAnnotations.test.ts`).
+ */
+import mermaid from "mermaid";
+
+function theme(): "dark" | "default" {
+  return document.body.classList.contains("vscode-dark") || document.body.classList.contains("vscode-high-contrast")
+    ? "dark"
+    : "default";
+}
+
+async function run(): Promise<void> {
+  const nodes = Array.from(document.querySelectorAll<HTMLElement>("pre.mermaid"));
+  if (nodes.length === 0) return;
+  mermaid.initialize({ startOnLoad: false, theme: theme(), securityLevel: "strict" });
+  try {
+    await mermaid.run({ nodes });
+  } catch {
+    // mermaid.run already marks failed nodes; ensure a visible chip per failure.
+    for (const n of nodes) {
+      if (!n.querySelector("svg")) n.setAttribute("data-cw-error", "true");
+    }
+  }
+}
+
+window.addEventListener("vscode.markdown.updateContent", () => void run());
+void run();
@@ -18,6 +18,28 @@
    "onStartupFinished"
  ],
  "contributes": {
+    "markdown.markdownItPlugins": true,
+    "markdown.previewStyles": [
+      "./media/preview-annotations.css"
+    ],
+    "markdown.previewScripts": [
+      "./out/media/preview-mermaid.js"
+    ],
+    "configuration": {
+      "title": "Cowriting",
+      "properties": {
+        "cowriting.liveProgress.revealOutput": {
+          "type": "boolean",
+          "default": true,
+          "description": "When Claude is editing, reveal the \"Cowriting: Claude\" output channel (without stealing focus) as soon as Claude starts producing text, so you can read the output as it streams."
+        },
+        "cowriting.annotations": {
+          "type": "boolean",
+          "default": true,
+          "description": "Show authorship + change annotations in the Markdown preview for coedited documents."
+        }
+      }
+    },
    "commands": [
      {
        "command": "cowriting.showClineSdkInfo",
@@ -45,13 +67,13 @@
        "category": "Cowriting"
      },
      {
-        "command": "cowriting.toggleAttribution",
-        "title": "Toggle Attribution",
+        "command": "cowriting.applyAgentEdit",
+        "title": "Apply Agent Edit (internal seam)",
        "category": "Cowriting"
      },
      {
-        "command": "cowriting.applyAgentEdit",
-        "title": "Apply Agent Edit (internal seam)",
+        "command": "cowriting.edit",
+        "title": "Ask Claude to Edit",
        "category": "Cowriting"
      },
      {
@@ -59,30 +81,80 @@
        "title": "Ask Claude to Edit Selection",
        "category": "Cowriting"
      },
-      {
-        "command": "cowriting.acceptProposal",
-        "title": "✓ Accept Proposal",
-        "category": "Cowriting"
-      },
-      {
-        "command": "cowriting.rejectProposal",
-        "title": "✗ Reject Proposal",
-        "category": "Cowriting"
-      },
      {
        "command": "cowriting.proposeAgentEdit",
        "title": "Propose Agent Edit (internal seam)",
        "category": "Cowriting"
      },
      {
-        "command": "cowriting.toggleDiffView",
-        "title": "Cowriting: Toggle Diff View",
+        "command": "cowriting.markReviewed",
+        "title": "Mark Changes as Reviewed",
+        "category": "Cowriting",
+        "icon": "$(check-all)"
+      },
+      {
+        "command": "cowriting.reviewChanges",
+        "title": "Review Changes",
+        "category": "Cowriting",
+        "icon": "$(git-compare)"
+      },
+      {
+        "command": "cowriting.openReviewPreview",
+        "title": "Open Cowriting Review Preview",
        "category": "Cowriting"
      },
      {
-        "command": "cowriting.pinDiffBaseline",
-        "title": "Cowriting: Pin Diff Baseline to Now",
+        "command": "cowriting.editDocument",
+        "title": "Ask Claude to Edit Document",
        "category": "Cowriting"
+      },
+      {
+        "command": "cowriting.acceptAllProposals",
+        "title": "Accept All Claude Proposals",
+        "category": "Cowriting"
+      },
+      {
+        "command": "cowriting.rejectAllProposals",
+        "title": "Reject All Claude Proposals",
+        "category": "Cowriting"
+      },
+      {
+        "command": "cowriting.proposalAcceptMenu",
+        "title": "Accept Claude Proposal",
+        "category": "Cowriting"
+      },
+      {
+        "command": "cowriting.proposalRejectMenu",
+        "title": "Reject Claude Proposal",
+        "category": "Cowriting"
+      },
+      {
+        "command": "cowriting.coeditDocument",
+        "title": "✦ Coedit this Document with Claude",
+        "category": "Cowriting"
+      },
+      {
+        "command": "cowriting.stopCoediting",
+        "title": "Stop editing with Claude",
+        "category": "Cowriting"
+      },
+      {
+        "command": "cowriting.askClaude",
+        "title": "Ask Claude",
+        "category": "Cowriting",
+        "icon": "$(sparkle)"
+      },
+      {
+        "command": "cowriting.makeThreadEdit",
+        "title": "✦ Make this edit",
+        "category": "Cowriting",
+        "icon": "$(sparkle)"
+      },
+      {
+        "command": "cowriting.toggleAnnotations",
+        "title": "Toggle Annotations",
+        "category": "Cowriting",
+        "icon": "$(eye)"
      }
    ],
    "menus": {
@@ -96,61 +168,168 @@
          "when": "false"
        },
        {
-          "command": "cowriting.acceptProposal",
+          "command": "cowriting.createThread",
+          "when": "editorHasSelection && (resourceScheme == file || resourceScheme == untitled) && cowriting.isCoediting"
+        },
+        {
+          "command": "cowriting.reviewChanges",
+          "when": "editorLangId == markdown && cowriting.isCoediting"
+        },
+        {
+          "command": "cowriting.markReviewed",
+          "when": "editorLangId == markdown && cowriting.isCoediting && cowriting.baselineMode == snapshot"
+        },
+        {
+          "command": "cowriting.edit",
+          "when": "editorLangId == markdown"
+        },
+        {
+          "command": "cowriting.editSelection",
          "when": "false"
        },
        {
-          "command": "cowriting.rejectProposal",
+          "command": "cowriting.editDocument",
          "when": "false"
+        },
+        {
+          "command": "cowriting.askClaude",
+          "when": "editorLangId == markdown && cowriting.isCoediting"
+        },
+        {
+          "command": "cowriting.toggleAnnotations",
+          "when": "editorLangId == markdown && cowriting.isCoediting"
+        },
+        {
+          "command": "cowriting.makeThreadEdit",
+          "when": "false"
+        },
+        {
+          "command": "cowriting.acceptAllProposals",
+          "when": "editorLangId == markdown && cowriting.isCoediting"
+        },
+        {
+          "command": "cowriting.rejectAllProposals",
+          "when": "editorLangId == markdown && cowriting.isCoediting"
+        },
+        {
+          "command": "cowriting.proposalAcceptMenu",
+          "when": "false"
+        },
+        {
+          "command": "cowriting.proposalRejectMenu",
+          "when": "false"
+        }
+      ],
+      "editor/title": [
+        {
+          "command": "cowriting.reviewChanges",
+          "when": "resourceLangId == markdown && cowriting.isCoediting",
+          "group": "navigation@1"
+        },
+        {
+          "command": "cowriting.askClaude",
+          "when": "resourceLangId == markdown && cowriting.isCoediting",
+          "group": "navigation@2"
+        },
+        {
+          "command": "cowriting.toggleAnnotations",
+          "when": "resourceLangId == markdown && cowriting.isCoediting",
+          "group": "navigation@3"
+        },
+        {
+          "command": "cowriting.markReviewed",
+          "when": "resourceLangId == markdown && cowriting.isCoediting && cowriting.baselineMode == snapshot",
+          "group": "navigation@4"
+        },
+        {
+          "command": "cowriting.openReviewPreview",
+          "when": "resourceLangId == markdown",
+          "group": "navigation@9"
+        }
+      ],
+      "editor/title/context": [
+        {
+          "command": "cowriting.edit",
+          "when": "resourceLangId == markdown && cowriting.isCoediting",
+          "group": "1_cowriting@1"
+        },
+        {
+          "command": "cowriting.openReviewPreview",
+          "when": "resourceLangId == markdown",
+          "group": "1_cowriting@3"
+        }
+      ],
+      "explorer/context": [
+        {
+          "command": "cowriting.openReviewPreview",
+          "when": "resourceLangId == markdown",
+          "group": "navigation@9"
        }
      ],
      "editor/context": [
        {
-          "command": "cowriting.editSelection",
-          "when": "editorHasSelection && resourceScheme == file",
+          "command": "cowriting.edit",
+          "when": "editorLangId == markdown && (resourceScheme == file || resourceScheme == untitled) && cowriting.isCoediting",
          "group": "1_cowriting@1"
        },
        {
          "command": "cowriting.createThread",
-          "when": "editorHasSelection && resourceScheme == file",
+          "when": "editorHasSelection && (resourceScheme == file || resourceScheme == untitled) && cowriting.isCoediting",
          "group": "1_cowriting@2"
+        },
+        {
+          "command": "cowriting.coeditDocument",
+          "when": "editorLangId == markdown && !cowriting.isCoediting",
+          "group": "1_cowriting@1"
+        },
+        {
+          "command": "cowriting.stopCoediting",
+          "when": "editorLangId == markdown && cowriting.isCoediting",
+          "group": "1_cowriting@9"
        }
      ],
      "comments/commentThread/context": [
        {
          "command": "cowriting.reply",
-          "group": "inline",
+          "group": "inline@1",
          "when": "commentController == cowriting.threads"
+        },
+        {
+          "command": "cowriting.makeThreadEdit",
+          "group": "inline@2",
+          "when": "commentController == cowriting.threads && commentThread =~ /\\boffer\\b/"
        }
      ],
      "comments/commentThread/title": [
+        {
+          "command": "cowriting.makeThreadEdit",
+          "group": "inline@1",
+          "when": "commentController == cowriting.threads && commentThread =~ /\\boffer\\b/"
+        },
        {
          "command": "cowriting.resolveThread",
          "group": "inline",
-          "when": "commentController == cowriting.threads && commentThread =~ /^open$/"
+          "when": "commentController == cowriting.threads && commentThread =~ /\\bopen\\b/"
        },
        {
          "command": "cowriting.reopenThread",
          "group": "inline",
-          "when": "commentController == cowriting.threads && commentThread =~ /^resolved$/"
-        },
-        {
-          "command": "cowriting.acceptProposal",
-          "group": "inline@1",
-          "when": "commentController == cowriting.proposals && commentThread =~ /^pending$/"
-        },
-        {
-          "command": "cowriting.rejectProposal",
-          "group": "inline@2",
-          "when": "commentController == cowriting.proposals"
+          "when": "commentController == cowriting.threads && commentThread =~ /\\bresolved\\b/"
        }
      ]
    },
    "keybindings": [
      {
-        "command": "cowriting.toggleDiffView",
-        "key": "ctrl+alt+d",
-        "when": "editorTextFocus"
+        "command": "cowriting.reviewChanges",
+        "key": "ctrl+alt+r",
+        "mac": "cmd+alt+r",
+        "when": "editorLangId == markdown && cowriting.isCoediting"
+      },
+      {
+        "command": "cowriting.edit",
+        "key": "ctrl+alt+e",
+        "mac": "cmd+alt+e",
+        "when": "editorTextFocus && editorLangId == markdown && cowriting.isCoediting"
      }
    ]
  },
@@ -159,15 +338,22 @@
    "watch": "node esbuild.mjs --watch",
    "typecheck": "tsc --noEmit",
    "test": "vitest run",
-    "pretest:e2e": "npm run build && tsc -p tsconfig.e2e.json",
+    "clean:e2e": "node -e \"require('fs').rmSync('out/test',{recursive:true,force:true})\"",
+    "pretest:e2e": "npm run build && npm run clean:e2e && tsc -p tsconfig.e2e.json",
    "test:e2e": "node ./out/test/e2e/runTest.js",
    "smoke:live": "npm run build && node scripts/smoke-live-turn.mjs",
+    "smoke:native": "npm run pretest:e2e && node scripts/smoke-native-loop.mjs",
    "vscode:prepublish": "node esbuild.mjs"
  },
  "dependencies": {
-    "@cline/sdk": "0.0.46"
+    "@cline/sdk": "0.0.46",
+    "diff": "^7.0.0",
+    "markdown-it": "^14.1.0",
+    "mermaid": "^11.0.0"
  },
  "devDependencies": {
+    "@types/diff": "^7.0.0",
+    "@types/markdown-it": "^14.1.0",
    "@types/mocha": "^10.0.7",
    "@types/node": "^22.0.0",
    "@types/vscode": "^1.90.0",
@@ -10,7 +10,15 @@ console.log(`instruction: ${instruction}`);
 console.log(`text:        ${text}`);
 try {
  const t0 = Date.now();
-  const result = await runEditTurn(instruction, text);
+  // #60: exercise the live-progress path against the real SDK — log each reduced
+  // snapshot so the smoke shows streaming/activity/tokens, not just the result.
+  const result = await runEditTurn(instruction, text, {
+    onProgress: (s) => {
+      const bits = [s.phase === "tool" ? `tool:${s.tool}` : s.phase, `${s.chars}c`];
+      if (s.tokens) bits.push(`${s.tokens}tok`);
+      console.log(`  progress: ${bits.join(" ")}`);
+    },
+  });
  console.log(`replacement: ${JSON.stringify(result.replacement)}`);
  console.log(`model:       ${result.model}`);
  console.log(`sessionId:   ${result.sessionId}`);
@@ -0,0 +1,67 @@
+#!/usr/bin/env node
+// Task 9 / spec §7.1 rung 3: the manual, real-@cline/sdk smoke for the native
+// comment loop -- mirrors scripts/smoke-live-turn.mjs's role for F3 (a live
+// login-backed check, deliberately NOT in CI), but this one drives the whole
+// EDH (comment -> reply -> offer -> proposal needs vscode's own comment/doc
+// APIs, unlike smoke-live-turn.mjs's bare module call) via
+// test/e2e/smoke/nativeLoop.ts, launched exactly like test/e2e/runTest.ts
+// launches the regular host-E2E suites -- just against a single smoke entry
+// point that is NOT swept into `npm run test:e2e` (it lives outside
+// test/e2e/suite, which is the only directory suite/index.ts's mocha glob
+// searches).
+//
+// Prereqs: `npm run pretest:e2e` (builds the extension + compiles
+// test/e2e/**). `npm run smoke:native` does this for you.
+//
+// Opens the real sandbox/ folder (the repo's EDH workspace root), writes a
+// throwaway doc, posts a real comment via ThreadController.createThreadOnSelection,
+// waits for the real @cline/sdk reply + offer, makes + accepts the edit, then
+// prints the created proposal ids and deletes the throwaway doc.
+import * as fs from "node:fs";
+import * as os from "node:os";
+import * as path from "node:path";
+import { fileURLToPath } from "node:url";
+import { runTests } from "@vscode/test-electron";
+
+const __dirname = path.dirname(fileURLToPath(import.meta.url));
+
+async function main() {
+  const projectRoot = path.resolve(__dirname, "..");
+  const extensionDevelopmentPath = projectRoot;
+  const extensionTestsPath = path.resolve(projectRoot, "out/test/e2e/smoke/nativeLoop");
+  const sandbox = path.resolve(projectRoot, "sandbox");
+
+  if (!fs.existsSync(`${extensionTestsPath}.js`)) {
+    console.error(
+      `smoke-native-loop: ${extensionTestsPath}.js is missing -- run "npm run pretest:e2e" first (or "npm run smoke:native", which does this for you).`,
+    );
+    process.exit(1);
+  }
+
+  // Same macOS UNIX-socket-length workaround as test/e2e/runTest.ts (a long
+  // worktree path + the default .vscode-test/user-data dir can blow past the
+  // ~103-char limit).
+  const userDataRoot = process.platform === "win32" ? os.tmpdir() : "/tmp";
+  const userDataDir = fs.mkdtempSync(path.join(userDataRoot, "cwud-smoke-"));
+
+  try {
+    await runTests({
+      extensionDevelopmentPath,
+      extensionTestsPath,
+      launchArgs: [sandbox, "--disable-extensions", "--user-data-dir", userDataDir],
+    });
+    console.log("smoke-native-loop: PASS");
+    process.exit(0);
+  } catch (err) {
+    console.error(
+      `smoke-native-loop: FAILED (expected when Claude Code is absent/signed out, or the sandbox/ folder is missing): ${
+        err instanceof Error ? err.message : String(err)
+      }`,
+    );
+    process.exit(1);
+  } finally {
+    fs.rmSync(userDataDir, { recursive: true, force: true });
+  }
+}
+
+main();
@@ -0,0 +1,140 @@
+# Session 0021.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-11T08-33 (PST)
+> End: 2026-06-11T11-52 (PST)
+> Type: planning-and-executing
+> Status: **FINALIZED**
+> Posture: autonomous (yolo)
+
+## Launch prompt
+
+> /goal Plan and execute Feature #21 (F7 rendered track-changes markdown
+> preview) from the graduated Solution Design at
+> content `specs/coauthoring-rendered-preview.md` — 4 slices per §7.2,
+> host-E2E tier, no LLM in CI.
+
+## Plan
+
+Plan-and-execute F7 (#21) from the graduated Solution Design
+`coauthoring-rendered-preview.md`, four slices per §7.2:
+
+- **SLICE-1** `trackChangesModel.ts` — vscode-free pure render engine: block
+  diff (LCS by normalized text) + word-level prose refinement + `markdown-it`
+  render + atomic code/mermaid blocks (INV-22/23). Add `markdown-it` + `diff`
+  deps. Vitest suite per §6.8.
+- **SLICE-2** Webview asset + build: `media/preview.ts` (receive HTML, run
+  `mermaid.run()`, theme CSS), esbuild second entry → `out/media/preview.js`
+  bundling mermaid; sealed CSP shell (INV-21).
+- **SLICE-3** `TrackChangesPreviewController` (vscode layer) + additive F6
+  `onDidChangeBaseline` event + `package.json` command/keybinding
+  (`cowriting.showTrackChangesPreview`, `ctrl+alt+r`, markdown-only) +
+  `CowritingApi` handle + test seam (`isOpen`/`getLastModel`); debounced live
+  update.
+- **SLICE-4** Host E2E (open/live/epoch/pin/non-markdown per §6.8) +
+  `docs/MANUAL-SMOKE-F7.md` + README F7 section.
+
+One PR; host-E2E tier (no browser/deploy stage); no LLM in CI.
+
+## Pre-state
+
+Clean `main`, in sync with `origin/main`. Epic #1 inner loop (F2–F6) fully
+shipped; F6 diff-view (`#17`/`#19`) was the immediate predecessor. F7's spec
+`coauthoring-rendered-preview.md` graduated in session 0020. The
+implementation plan was written just-in-time from that spec
+(`docs/superpowers/plans/2026-06-11-f7-rendered-track-changes-preview.md`).
+
+## Turn-by-turn arc
+
+1. **Init.** Classified as planning-and-executing; claimed session **0021**
+   (no other sessions in flight). Verified baseline (fast-forwarded `main` 1
+   commit), read the F7 spec, resolved content/roadmap repos.
+2. **Plan.** Wrote the 12-task implementation plan with `superpowers:writing-plans`
+   (engine → webview → controller → wiring → E2E → docs → verify), self-reviewed
+   against the spec (full INV-20..23 + PUC coverage). Committed on branch
+   `f7-rendered-preview`.
+3. **Execute (TDD, `superpowers:executing-plans`).** Twelve tasks, each
+   test-first where applicable, committed individually:
+   - T1 deps (`markdown-it`, `diff`, `mermaid` + types).
+   - T2–T5 `src/trackChangesModel.ts` (pure engine): `splitBlocks` → `diffBlocks`
+     (LCS via jsdiff `diffArrays`, atomic fences INV-23) → `renderTrackChanges`
+     (markdown-it + custom mermaid fence rule + word-level prose `<ins>`/`<del>`)
+     → injectable per-block renderer + error-chip fallback (PUC-6). **18 vitest
+     cases.**
+   - T6 `media/preview.ts` + `media/preview.css` + esbuild **second entry**
+     (browser IIFE bundling mermaid → `out/media/preview.js`; CSS imported so
+     esbuild emits `out/media/preview.css`). Host bundle stayed ~68kb; webview
+     bundle ~7MB (mermaid, isolated as intended).
+   - T7 additive `DiffViewController.onDidChangeBaseline` event.
+   - T8 `src/trackChangesPreview.ts` — `TrackChangesPreviewController` (sealed
+     CSP+nonce webview, one panel/markdown doc, debounced live refresh, reuses F6
+     baseline, no persistence INV-20, test seam `isOpen`/`getLastModel`).
+   - T9 wired into `extension.ts` (workspace-independent like F6) + `CowritingApi`
+     + `package.json` command/keybinding (`ctrl+alt+r`, markdown-only).
+   - T10 host E2E `test/e2e/suite/trackChangesPreview.test.ts` (5 cases:
+     open/PUC-1, type/PUC-2, propose+accept epoch advance/PUC-3, pin/PUC-4,
+     non-markdown guard/PUC-6) + fixtures `preview.md` + `notes.txt`.
+   - T11 `docs/MANUAL-SMOKE-F7.md` + README F7 section.
+   - T12 full verification.
+4. **Verify.** typecheck clean; **122 unit** passing; **33 (workspace) + 4
+   (no-workspace) E2E** passing (F7's 5 among them); build produces all artifacts;
+   working tree clean (INV-20 — preview wrote nothing to repo/sidecar).
+5. **Ship.** Pushed branch, created **PR #23** via Gitea API, merged (merge
+   commit), synced `main`, deleted branch local + remote, re-verified `main`
+   green post-merge.
+
+## Cut state
+
+F7 (#21) shipped to `main` (PR #23, merge commit `ce01ef8`). Plan archived to the
+content repo `plans/` (`0a52159`). `main` clean and in sync. Deferred follow-up
+**#22** (intra-diagram mermaid diffing) remains open.
+
+## Deferred decisions
+
+Autonomous-mode low-confidence calls made this session (none blocked progress;
+all cheap to revisit):
+
+1. **CSP `style-src 'unsafe-inline'`** in the webview shell. Mermaid injects
+   `<style>` tags into the DOM at runtime, which a nonce/hash can't cover.
+   *Chosen:* allow inline styles while keeping **scripts** strictly nonce-gated
+   and local-only (no remote/CDN script source) — INV-21's network seal is
+   intact. *Alternative:* hash every injected style (impractical for runtime
+   injection). Standard practice for mermaid-in-webview, but flagging the
+   relaxation since it's security-adjacent.
+2. **Block-key normalization lowercases + collapses whitespace** for diff
+   matching. *Consequence:* a case-only or whitespace-only edit to a block won't
+   register as a `changed` block. *Chosen:* reduce diff noise from reflow/casing.
+   *Alternative:* case/whitespace-sensitive keys (more literal, noisier). Minor
+   fidelity tradeoff in a pure-preview view.
+3. **PR merged as a merge commit** (not squash), matching recent repo history
+   (`Merge pull request …` style). The 12 conventional commits are preserved.
+4. **CSS delivered via `import "./preview.css"` from the webview entry** so
+   esbuild emits the sibling `out/media/preview.css` (single `localResourceRoot`,
+   one build mechanism) rather than a separate copy step.
+
+## Operator plate
+
+- F7 is live: `Ctrl+Alt+R` / "Cowriting: Open Track-Changes Preview" on a
+  markdown doc. Run the **manual smoke** (`docs/MANUAL-SMOKE-F7.md`) once to
+  verify the webview's *visual* rendering (mermaid diagrams, theming) — that
+  layer is intentionally not asserted in the sealed-sandbox E2E.
+- **Deployment pipeline (§9):** this app is a **non-shippable VS Code extension
+  POC** (spec §7.3 — no marketplace publish). There is **no deployable UI app /
+  no PPE or prod stage** for it; the app's required pipeline tier is **host E2E**,
+  which ran green. So "completion" = merged + host-E2E green + manual smoke, not a
+  flotilla deploy. Nothing was skipped — there is no cloud stage to run.
+
+## Next session
+
+The open follow-up is **#22 (intra-diagram mermaid diffing)**, `type/task`. The
+F7 spec §6.7 explicitly flagged it as "a large unscoped design" (diff at the
+source / parsed-graph / SVG level; layout reflow) — so although typed a task, it
+is design-heavy and the session may want to scope the approach first.
+
+```
+/goal Plan and execute #22 (F7 intra-diagram mermaid diffing) — extend the F7 track-changes preview beyond the whole-diagram "changed" badge to node/edge-level diffing; spec §6.7 flags it as unscoped (source vs parsed-graph vs SVG, layout reflow), so scope the approach first. Builds on src/trackChangesModel.ts (atomic mermaid blocks, INV-23) + the F7 webview; host-E2E tier, no LLM in CI.
+```
+
+Read first: memory `f7-rendered-preview-shipped.md` (what shipped + the open
+follow-ups) and the F7 spec `coauthoring-rendered-preview.md` §6.7 (the deferred
+mermaid-diff decision).
@@ -0,0 +1,103 @@
+# Session 0022.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-11T12-09 (PST)
+> End: 2026-06-11T12-14 (PST)
+> Type: planning-and-executing
+> Status: **FINALIZED**
+> Posture: autonomous (yolo)
+
+## Launch prompt
+
+> Fix cowriting.editSelection guard: accurate per-condition warnings (no editor /
+> no selection / unsaved / outside-workspace) instead of one misleading "select
+> some text" message, and make the workspace-membership check use a
+> path-separator boundary (fix the startsWith prefix-collision latent bug).
+> Reported during F7 manual testing: editSelection on a content-repo file
+> (outside the EDH sandbox root) warned "highlight text" though text was selected.
+
+## Plan
+
+Trivial bugfix — no formal `writing-plans` artifact (planning-and-executing
+addendum: a trivial fix executed without a plan skips plan archival). Done via
+`systematic-debugging` (root cause → failing test → fix → verify).
+
+## Pre-state
+
+Clean `main`, in sync. Followed immediately on session 0021 (F7 shipped). Reported
+during F7 manual smoke.
+
+## Root cause (systematic-debugging Phase 1)
+
+`cowriting.editSelection`'s guard fired its "select some text in a workspace
+document first" warning, but the user had text selected. The guard collapsed four
+distinct conditions into one message:
+
+```
+!editor || selection.isEmpty || scheme !== "file" || !fsPath.startsWith(root)
+```
+
+Invoked via the editor **context menu**, whose `when: editorHasSelection &&
+resourceScheme == file` already guarantees conditions 1–3 — so by elimination the
+firing condition was **(4)**, the workspace-membership check. Confirmed
+empirically: the EDH opens `${workspaceFolder}/sandbox` as its root
+(`.vscode/launch.json`), but the file was
+`vscode-cowriting-plugin-content/issues/diff-view-toggle.md` — a *different repo*,
+outside `sandbox/`. `file.startsWith(root)` → `false` → condition 4 fired, with the
+wrong (selection-blaming) message. Also surfaced a **latent prefix-collision**: a
+plain `startsWith(root)` falsely matches a sibling whose name prefixes the root
+(`vscode-cowriting-plugin-content` vs `vscode-cowriting-plugin`).
+
+## Turn-by-turn arc
+
+1. **Init.** Claimed session **0022** (no other in flight); clean baseline.
+2. **Debug.** `systematic-debugging`; asked two diagnostic questions + the file
+   path; confirmed root cause empirically with a node one-liner (Phase 3).
+3. **Fix (TDD, Phase 4).** New pure, vscode-free `src/workspacePath.ts`:
+   `isUnderRoot(fsPath, root)` (separator-bounded membership) +
+   `selectionRejection(ctx)` (one message per condition). Failing test first
+   (`test/workspacePath.test.ts`, 8 cases incl. the reported bug + the
+   prefix-collision case), then implementation → green.
+4. **Wire.** Replaced `startsWith(root)` at all **five** sites: `extension.ts`
+   (editSelection guard rewritten to use `selectionRejection`; `renderIfOpen`) +
+   thread/attribution/proposal controllers' membership checks.
+5. **Verify.** typecheck clean; **130 unit** (122 + 8); **33 + 4 E2E** (the
+   existing F2–F7 suites exercise the membership checks through real workspace
+   fixtures = regression guard for `isUnderRoot`).
+6. **Ship.** Branch `fix-editselection-guard` → **PR #24** → merged (merge
+   commit `0525c40`); `main` synced; branch deleted local + remote.
+
+## Cut state
+
+Fix shipped to `main` (PR #24). `main` clean and in sync. No plan artifact (trivial
+fix). The deferred F7 follow-up **#22** (intra-diagram mermaid diffing) remains the
+open next item.
+
+## Deferred decisions
+
+No low-confidence calls this session — the root cause was confirmed empirically
+before any code change, and the fix is a direct, test-covered correction.
+
+## Operator plate
+
+- **To test "Ask Claude to Edit Selection," use a markdown file inside the EDH
+  workspace folder** (`.../vscode-cowriting-plugin/sandbox/`). F3/F4 (threads /
+  attribution / editSelection) require a saved `file:` under the workspace root,
+  because they persist a `.threads/` sidecar beside it; F6/F7 work on any file.
+  This inconsistency is recorded in memory `editselection-workspace-membership.md`.
+- The warning now names the real reason (no editor / no selection / unsaved /
+  outside workspace) instead of always "select some text."
+
+**Deployment pipeline (§9):** non-shippable VS Code extension POC — no PPE/prod
+stage; host-E2E is the tier and it's green. Nothing skipped.
+
+## Next session
+
+Unchanged by this fix — the open item is **#22 (intra-diagram mermaid diffing)**.
+
+```
+/goal Plan and execute #22 (F7 intra-diagram mermaid diffing) — extend the F7 track-changes preview beyond the whole-diagram "changed" badge to node/edge-level diffing; spec §6.7 flags it as unscoped (source vs parsed-graph vs SVG, layout reflow), so scope the approach first. Builds on src/trackChangesModel.ts (atomic mermaid blocks, INV-23) + the F7 webview; host-E2E tier, no LLM in CI.
+```
+
+Read first: memory `f7-rendered-preview-shipped.md`, `editselection-workspace-membership.md`,
+and F7 spec `coauthoring-rendered-preview.md` §6.7.
@@ -0,0 +1,79 @@
+# Session 0023.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-11T12-28 (PST)
+> End: 2026-06-11T12-30 (PST)
+> Type: capture (tracked-lite)
+> Status: **FINALIZED**
+> Posture: autonomous (yolo)
+
+## Launch prompt
+
+> Capture a Feature on the tracker: extend the authoring flow (F2/F3/F4 — threads /
+> attribution / Ask Claude to Edit Selection) to files OUTSIDE the workspace folder,
+> persisting the sidecar in VS Code global storage keyed by URI hash (like F6's
+> baseline) for out-of-workspace/untitled files. Hybrid model recommended. Origin:
+> F7 manual testing (sessions 0021/0022); operator chose global storage.
+
+## What this captured
+
+One **Feature** issue, filed to the tracker:
+
+- **#25 — Out-of-workspace authoring (global-storage sidecar fallback)**
+  (`type/feature`, `priority/P2`):
+  https://git.benstull.org/benstull/vscode-cowriting-plugin/issues/25
+
+The ask: extend F2/F3/F4 (threads / attribution / "Ask Claude to Edit Selection")
+to files outside the workspace folder + untitled buffers, via a **hybrid**
+persistence model — in-workspace `file:` docs keep the committable git-native
+`.threads/` sidecar (INV-2 unchanged); out-of-workspace / untitled docs fall back to
+a **global-storage sidecar keyed by URI hash** (like F6's baseline, INV-19). The
+issue records the F5 cross-rung non-shareability of global-storage artifacts and
+the URI-rename orphaning caveat, and flags that — being a persistence-model change
+— it needs a **Solution Design before code** (handbook §3.4 R3).
+
+## Arc
+
+1. **Origin (sessions 0021/0022).** During F7 manual testing, "Ask Claude to Edit
+   Selection" was refused on an out-of-workspace content-repo file. Session 0022
+   (PR #24) fixed the misleading message + the `startsWith` prefix-collision. The
+   operator then asked why authoring (unlike F6/F7) doesn't work on any file.
+2. **Design conversation.** Established the asymmetry (diff/preview persist outside
+   the repo → any file; authoring persists a repo-relative `.threads/` sidecar →
+   workspace-only). Operator chose the **global-storage** fork over co-locate /
+   keep-as-is, and chose to **capture it for later** ("we'll take it from there").
+3. **Capture (this session).** Claimed tracked-lite session 0023; resolved the
+   content repo; authored the §5 issue draft
+   (`issues/out-of-workspace-authoring.md`); ensured labels; filed **#25**.
+
+## Notes
+
+- **Token:** the capture scripts' default Keychain entry
+  (`wgl-gitea-issues-readwrite-token`) is absent for `git.benstull.org` (401);
+  pointed `WGL_CAPTURE_TOKEN_SERVICE` at the existing
+  `wgl-gitea-token-git.benstull.org` (write:repository) to ensure labels + file the
+  issue. (A dedicated issue-scoped token for this host would let the default work.)
+- **Draft (INV-8):** `vscode-cowriting-plugin-content/issues/out-of-workspace-authoring.md`
+  is left **uncommitted** in the content-repo working tree — the Author's to commit/
+  publish or discard (this skill never pushes the content repo). Several earlier
+  capture drafts are likewise uncommitted there.
+
+## Deferred decisions
+
+No low-confidence calls — the operator directly chose the global-storage approach
+and directed the capture. The hybrid-vs-uniform fork is recorded in the issue as the
+key open design decision for the eventual Solution Design (not decided here).
+
+## Next session
+
+Two open items; pick per priority:
+
+```
+/goal Brainstorm and write the Solution Design for #25 (out-of-workspace authoring) — settle the hybrid persistence model (in-workspace .threads/ sidecar vs global-storage fallback for out-of-workspace/untitled), the F5 cross-rung non-shareability, and the URI-rename orphaning caveat; per the issue body.
+```
+
+(The earlier F7 follow-up **#22** — intra-diagram mermaid diffing — also remains
+open; sequence #22 vs #25 by priority. #25 is `priority/P2`.)
+
+Read first: issue #25, memory `editselection-workspace-membership.md`, and the F2–F6
+specs for the sidecar/baseline persistence precedents.
@@ -0,0 +1,54 @@
+# Session 0024.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-11T12-36 (PST) · End: 2026-06-11T12-50 (PST, approximate)
+> Type: brainstorming
+> Driver: Ben Stull (with Claude)
+> Status: **FINALIZED (adopted)** — see the note below.
+
+## Adoption note
+
+This session ended **without finalizing** (it was left `--INPROGRESS`), so its
+original turn-by-turn conversation was not preserved. It was **adopted and
+finalized by session 0026's driver** on 2026-06-11 (per the finalize-on-goal /
+adopt-and-finalize policy). The end time above is approximate — the session ended
+sometime before session 0025 claimed at 12-51. The content below is **reconstructed**
+from the artifact this session produced (the F8 Solution Design) and the project
+memory written at the time, not from a live transcript.
+
+## Launch prompt
+
+```
+Brainstorm and write the Solution Design for #25 (out-of-workspace authoring) —
+settle the hybrid persistence model, the F5 cross-rung non-shareability, and the
+URI-rename orphaning caveat; per the issue body.
+```
+
+## Output (the reviewed artifact)
+
+The session produced and graduated the **F8 Solution Design**,
+`specs/coauthoring-out-of-workspace.md` (submitted to the content repo's `specs/`
+collection during this adopted finalize — it had not been submitted when the
+session ended). Forks settled (per the spec §9 + memory):
+
+- **Hybrid persistence (load-bearing):** a `SidecarStore` interface + `SidecarRouter`
+  façade with two impls — repo-rooted `CoauthorStore` (in-workspace `file:` →
+  committable `.threads/`, INV-2 byte-for-byte) and a new `GlobalSidecarStore`
+  (out-of-workspace `file:` + `untitled:` → global storage by `sha256(uri)`,
+  INV-19). Routed per-document by `#24`'s `isUnderRoot`. → INV-24.
+- **F5 cross-rung non-shareability:** a global artifact isn't a committed file, so
+  it is single-rung by construction. → INV-25.
+- **URI-rename orphaning:** accepted & documented (inherent to URI-hash keying).
+- **Document key** = repo-relative path in-workspace / URI string otherwise;
+  shared `isAuthorable` gate; untitled in-memory only (no on-save migration in v1).
+
+## Downstream
+
+F8 was implemented from this design in **session 0025** (PR #26, shipped to `main`)
+and the diff/preview authorship follow-up shaped F9 (session 0026, PR #27). See the
+project memory `f8-out-of-workspace-spec-graduated` / `f8-out-of-workspace-shipped`.
+
+## Deferred decisions
+
+None recorded (the placeholder's section was empty; this is an adopted finalize, so
+no new low-confidence calls were made here beyond the approximate end time).
@@ -0,0 +1,104 @@
+# Session 0025.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-11T12-51 (PST) · End: 2026-06-11T13-21 (PST)
+> Type: planning-and-executing
+> Driver: Ben Stull (with Claude / Opus 4.8)
+> Status: **FINALIZED**
+
+## Launch prompt
+
+`/goal next` → the session gate resolved the stored next-goal from memory:
+**plan-and-execute F8 (out-of-workspace authoring)** from the graduated spec
+`specs/coauthoring-out-of-workspace.md` (Feature #25, brainstorming session 0024).
+
+## Pre-state
+
+- `main` clean but 5 commits behind `origin/main` (session-history transcripts
+  0022/0023/0024 + the 0025 placeholder); fast-forwarded.
+- An in-flight predecessor: **session 0024** (brainstorming, specced F8) still
+  `--INPROGRESS` in `sessions/0024/`. Sequential handoff (spec → code, no
+  code-surface overlap) → noted and proceeded in autonomous posture.
+- The F8 spec sat untracked at `specs/coauthoring-out-of-workspace.md` (the
+  brainstorming artifact); used as the read reference, left untracked (specs
+  don't belong in a code PR).
+- Baseline: 130 unit tests green.
+
+## Plan
+
+Wrote the implementation plan with `superpowers:writing-plans` →
+`docs/superpowers/plans/2026-06-11-f8-out-of-workspace.md` (12 TDD tasks across the
+spec's four slices). Executed inline (sequential dependencies, shared files) with
+`superpowers:test-driven-development`.
+
+## Turn-by-turn arc
+
+1. **Classify + claim.** `/goal next` → planning-and-executing → `wgl-planning-and-executing`
+   claimed session **0025**, fast-forwarded `main`, confirmed the resume pointer.
+2. **Read the spec + the whole authoring surface** (store, baselineStore,
+   workspacePath, the three controllers, extension wiring, versionGuard, model,
+   package.json, e2e harness) before planning.
+3. **SLICE-1** (Tasks 1–5): `SidecarStore` interface; `GlobalSidecarStore`
+   (sha256-keyed disk for out-of-folder `file:`, in-memory for `untitled:`;
+   INV-16 throw + anchor prune); `SidecarRouter` (`keyOf` + per-key routing);
+   `isAuthorable` + widened `selectionRejection`. TDD per file; 18 new unit tests.
+   - One hiccup: a `**/` glob inside a JSDoc block comment contained `*/` and
+     closed the comment early (esbuild "Unexpected ."). Reworded to `.threads/`.
+4. **SLICE-2** (Tasks 6–7): re-pointed `VersionGuard` + the three controllers
+   from `CoauthorStore` onto the router; gate → `isAuthorable`; key via
+   `store.keyOf(docIdentity(...))`; `rootDir` → `string | undefined` (git email
+   omitted when no root). `CoauthorStore` left untouched.
+5. **SLICE-3** (Task 8): wired the router in `extension.ts`; removed the no-root
+   early-return + command stubs so authoring is live folder-less (F6 #19
+   precedent); `renderIfOpen` → `isAuthorable`; widened the editor-context menus
+   to `untitled`; exported `sidecarRouter` on `CowritingApi`. 150 unit tests green.
+6. **SLICE-4** (Tasks 9–12): host E2E for out-of-folder file + untitled +
+   in-workspace byte-for-byte regression; rewrote the no-workspace suite
+   (authoring now real folder-less); `docs/MANUAL-SMOKE-F8.md` + README F8 note.
+   Full E2E: **36 passing** with-workspace + **5 passing** no-workspace.
+7. **Self-review:** `model.ts` / `mergeArtifacts.ts` / `store.ts` confirmed
+   untouched (INV-2/24/25 by construction).
+8. **Ship:** PR **#26** (`f8-out-of-workspace` → `main`) created + merged via the
+   Gitea API; branch deleted; `main` fast-forwarded; 150 unit tests green on the
+   merged tree. Issue **#25 closed**. Plan archived to the content repo
+   `plans/2026-06-11-f8-out-of-workspace.md`.
+
+## Cut state
+
+- F8 merged to `main` (PR #26); issue #25 closed; plan archived.
+- Tests: 150 unit + 41 host E2E green.
+- VS Code extension → no PPE/prod pipeline stage (no deployable UI); spec §7.3
+  declares it non-shippable. Done = merge + tests green.
+
+## Deferred decisions
+
+Autonomous-mode low-confidence calls surfaced for the operator:
+
+- **Controllers typed `SidecarRouter` (concrete), not the bare `SidecarStore`
+  interface.** The spec (§6.4) wrote `store: SidecarStore` + `store.keyOf`, but
+  `keyOf`/`sidecarPath` live on the router, not the 4-method storage interface
+  (which `CoauthorStore`/`GlobalSidecarStore` implement). Typing controllers to
+  the concrete router is the faithful, fully-typed realization. Least-churn,
+  revisitable.
+- **Editor-context menus widened to `untitled`** (`resourceScheme == file ||
+  resourceScheme == untitled`) so the right-click affordance matches the widened
+  gate — a small UX call not spelled out in the spec.
+- **Session 0024 left unfinalized** (out of this session's scope). The
+  brainstorming predecessor that specced F8 is still `--INPROGRESS` in
+  `sessions/0024/`; it needs an adopt-and-finalize pass. Its spec artifact sits
+  untracked at `specs/coauthoring-out-of-workspace.md`, and the spec home is
+  ambiguous (app.json lists `specs` under the main repo; `submit-spec.sh` targets
+  the content repo). Flagged, not resolved.
+- **Issue hygiene** (prior sessions): #21 (F7) still open though F7 shipped (PR
+  #23); Epic #1 still open though memory records it fully shipped. Not touched.
+
+## Next-session prompt
+
+```
+/goal plan-and-execute #22 (F7 intra-diagram mermaid diffing — node/edge-level diff beyond the whole-diagram "changed" badge), per Feature #22
+```
+
+Read [[f8-out-of-workspace-shipped]] for the SidecarStore/router surface and the
+open F8 follow-ups (untitled→saved migration, re-key/recover gesture). Consider an
+adopt-and-finalize pass on session **0024** first (it's a one-step finalize), and
+the #21/#1 issue-status hygiene.
@@ -0,0 +1,87 @@
+# Session 0026.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-11T14-29 (PST) · End: 2026-06-11T14-42 (PST)
+> Type: planning-and-executing
+> Driver: Ben Stull (with Claude / Opus 4.8)
+> Status: **FINALIZED**
+
+## Launch prompt
+
+Continuation of the same conversation as session 0025 (which shipped F8). The
+operator reported friction: **"It's not showing the Claude-composed annotations in
+the preview."** Debugging clarified it was a *feature gap*, not a regression — the
+F7 preview is author-agnostic by design. This session designed and shipped the fix:
+**F9 — an Authorship mode in the rendered preview**.
+
+## Pre-state
+
+- F8 shipped (session 0025, PR #26); `main` clean.
+- Debugging (systematic-debugging) found: `renderTrackChanges` diffs only baseline
+  vs buffer (no attribution), and the F6 baseline advances on machine-landing
+  (INV-18), so Claude's accepted text reads as "unchanged". → an authorship axis is
+  needed, read from F3.
+- `AskUserQuestion` settled the design: a **toggle** (changes ⟷ authorship),
+  **inline char-precise** marking, **both authors** colored (Claude blue / human
+  green), **block-level** badges for code/mermaid fences, **segmented header toggle**.
+
+## Plan
+
+Brainstormed the design (`superpowers:brainstorming`) → spec
+`docs/superpowers/specs/2026-06-11-authorship-preview-design.md` (INV-26..28).
+After operator approval, wrote the plan (`superpowers:writing-plans`) →
+`docs/superpowers/plans/2026-06-11-f9-authorship-preview.md` (7 TDD tasks) and
+executed inline (`superpowers:test-driven-development`).
+
+## Turn-by-turn arc
+
+1. **Debug (read-only):** read the full F7 + F3 stack; identified the feature gap +
+   baseline tension. Confirmed with the operator via AskUserQuestion ("mark what
+   Claude wrote" → attribution-aware preview).
+2. **Brainstorm:** four design decisions locked (axis=toggle, granularity=inline,
+   scope=both authors, fences=block badge, toggle=segmented header). Wrote +
+   committed the spec; operator approved.
+3. **Claim 0026** (planning-and-executing) before implementation.
+4. **SLICE-1/2 (Tasks 1–2):** `splitBlocksWithRanges` (block source offsets) + pure
+   `renderAuthorship` — PUA sentinel injection through markdown-it, atomic fences,
+   adjacent-span ordering. 29 model unit tests.
+5. **SLICE-2 (Tasks 3–4):** `AttributionController.spansFor`; preview gains the
+   attribution dep + per-panel mode + `setMode`; `extension.ts` reorder.
+6. **SLICE-3 (Task 5):** webview segmented toggle + legend + `.cw-by-*` CSS.
+7. **SLICE-4 (Tasks 6–7):** host E2E (authorship marks Claude's landed span +
+   track-changes regression) + manual smoke + README.
+8. **Ship:** PR **#27** merged to `main`; `main` synced; 161 unit + 37/5 host E2E
+   green. Plan archived + spec graduated to the content repo.
+
+Notable hiccup: the PUA sentinels are invisible characters — they round-tripped
+correctly through the Write tool (verified via `cat -v`), but a couple of test
+regexes and a doc literal needed explicit `..E003` escapes for clarity.
+
+## Cut state
+
+- F9 merged to `main` (PR #27). Spec graduated + plan archived (content repo).
+- Tests: 161 unit + 37 (with-ws) + 5 (no-ws) host E2E green.
+- Core files (`model.ts`, `store.ts`, `mergeArtifacts.ts`, `diffViewController.ts`,
+  `baselineStore.ts`) untouched; track-changes mode behavior unchanged.
+- VS Code extension → no PPE/prod stage; done = merge + tests green.
+
+## Deferred decisions
+
+- **F9 typed/built without a captured tracker issue** — it came from a conversational
+  bug report; PR #27 carries the record. A retro Feature issue would tidy the tracker.
+- **Inline (not block-level) authorship marking** — the operator chose inline
+  precision; the residual risk (a sentinel adjacent to `**`/`_` perturbing that
+  span's markdown) is rare, fails soft to a per-block chip, and hardening is deferred.
+- **Session 0024 still unfinalized** — the finalize-on-goal hook flagged it twice;
+  the operator deferred the cleanup (declined a finalize detour). Still an open
+  `--INPROGRESS` needing an adopt-and-finalize pass. Plus #21/#1 issue-status hygiene.
+
+## Next-session prompt
+
+```
+/goal plan-and-execute #22 (F7 intra-diagram mermaid diffing — node/edge-level diff beyond the whole-diagram "changed" badge), per Feature #22
+```
+
+Read [[f9-authorship-preview-shipped]] for the render/wiring surface. Consider
+finalizing session **0024** first (a one-step adopt-and-finalize) and the #21/#1
+issue hygiene.
@@ -0,0 +1,121 @@
+# Session 0027.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-11T16-16 (PST)
+> End: 2026-06-11T20-33 (PST)
+> Type: planning-and-executing
+> Posture: autonomous (yolo)
+> Status: **FINALIZED**
+
+## Launch prompt
+
+```
+plan-and-execute #22 (F7 intra-diagram mermaid diffing — node/edge-level diff
+beyond the whole-diagram "changed" badge), per Feature #22.
+```
+
+## Outcome (one line)
+
+Task **#22** — intra-diagram mermaid diffing — shipped to `main` via **PR #28**
+(issue auto-closed); 3 new pure host modules + render wire-in; **189 unit + 38/5
+host E2E** green; design §11 (INV-29..31) + plan archived to the content repo.
+
+## Pre-state
+
+- `main` clean & pushed; Epic #1 + F2–F9 shipped (last: F9 authorship preview,
+  PR #27, session 0026). #22 (F7's deferred intra-diagram task, `type/task`) open.
+- F7 render engine in place: `src/trackChangesModel.ts` (pure block diff, mermaid
+  fences **atomic** per INV-23, whole-diagram "changed" badge) + sealed webview
+  (`media/preview.ts` runs `mermaid.run()`) + `TrackChangesPreviewController`.
+- Carried-in loose ends: session 0024 unfinalized; a stray untracked
+  `specs/coauthoring-out-of-workspace.md` in the code repo working tree.
+
+## Arc (turn by turn)
+
+1. **Init.** Claimed session 0027 (no in-flight sessions). Verified clean pushed
+   `main`. Read memory resume pointer (Next /goal = #22).
+2. **Gate (§4.3).** Confirmed #22 is `type/task` — a leaf, eligible
+   planning-and-executing anchor (R2); no separate Solution Design required.
+3. **Brainstorming (superpowers).** Explored the F7 architecture to ground the
+   design. Resolved the four forks the issue flagged:
+   - **Diff level → parsed-graph** (source-text defeats a *rendered* preview; SVG
+     diff is layout-brittle — §6.7 already rejected it). Parse → diff → re-emit the
+     current source augmented with mermaid's *own* styling directives, keeping the
+     host pure (INV-22) and the webview unchanged (INV-21).
+   - **Layout reflow → accepted** (no position pinning; mermaid exposes none).
+   - Asked the operator the two genuinely-open forks via AskUserQuestion:
+     - **Removed elements → "ghost in place"** (faded/dashed, re-injected at
+       baseline position).
+     - **Diagram scope → "flowchart + sequence"** (broader than my flowchart-only
+       recommendation); all else → v1 badge fallback.
+   - Surfaced a real mermaid constraint: sequence diagrams have **no per-message
+     color hook** → use `rect rgb(...)` tinted bands (the only option).
+   - Wrote the design as **§11** of the F7 spec (content repo) + INV-29..31.
+4. **Plan (superpowers:writing-plans).** 9 TDD tasks → `docs/superpowers/plans/2026-06-11-f7-intra-diagram-mermaid-diff.md`.
+5. **Execute (superpowers:executing-plans, inline).** Feature branch
+   `f7.1-intra-diagram-mermaid-diff`; one commit per task:
+   - `mermaidDiff.ts` (dispatch + `detectDiagramType` + total try/catch fallback;
+     owns `CW_COLORS`).
+   - `mermaidFlowchartDiff.ts` — `parseFlowchart` + `diffFlowchart` →
+     `classDef`/`class`/`linkStyle`, ghost nodes/edges re-injected.
+   - `mermaidSequenceDiff.ts` — `parseSequence` + `diffSequence` (LCS via jsdiff)
+     → `rect` tinted runs, removed messages/participants ghosted.
+   - Wired into `renderOp`'s changed-atomic-mermaid branch (`mermaidFenceBody`
+     helper + `MERMAID_LEGEND`); legend CSS; `renderHtmlFor` E2E seam.
+   - Updated one pre-existing unit test that asserted the *old* v1 badge for a
+     changed flowchart (now augments — exactly what #22 changes).
+6. **Verify & ship.** typecheck clean, 189 unit, 38/5 E2E. PR #28 opened on Gitea
+   (needed the repo-scoped keychain token — the issues-only token lacked
+   `read:repository` scope), merged to `main`, branch deleted, #22 auto-closed.
+7. **Spec reconcile.** The local content clone was 6 commits stale (remote had the
+   *graduated* F7 spec + F8/F9). Aborted a conflicting rebase, reset local to
+   origin, re-applied §11 cleanly onto the current spec, pushed (`1c2844d`).
+8. **Operator Q (mid-finalize):** asked about the macOS "VS Code would like to
+   access data from other apps" prompt → explained it's the TCC cross-app-data
+   gate triggered by my Keychain reads (`security` CLI fetching the Gitea token),
+   not the file edits; gave Allow/Don't-Allow trade-offs.
+9. **Finalize.** Archived the plan to the content repo `plans/` (`6b14de9`);
+   updated memory; published this transcript.
+
+## Cut state
+
+- `main` (code): PR #28 merged (`06d4f87`), clean & pushed. Only the carried-in
+  stray `specs/` dir remains untracked (not this session's).
+- Content repo: F7 spec §11 (`1c2844d`) + archived plan (`6b14de9`) pushed.
+  Untracked `issues/*.md` are pre-existing capture drafts, untouched.
+- Issue #22: **closed**. No open issues remain.
+
+## Deployment pipeline (§9)
+
+**No pipeline stage applies.** This app is a VS Code extension with **no
+flotilla/cloud infra** (`app.json`: code/specs/roadmap/sessions repos only).
+localhost + E2E (host) ran green; there is no PPE/prod deploy stage to gate.
+Done = merge + green tests.
+
+## Deferred decisions
+
+_Autonomous-mode low-confidence calls; none material this session._
+
+- The two product-flavored forks (removed-element treatment, diagram scope) were
+  **not** auto-decided — they were put to the operator via AskUserQuestion and
+  answered (ghost-in-place; flowchart+sequence). Everything else had a clearly
+  preferred answer and was decided autonomously.
+- Updated one pre-existing unit test to the new behavior (judgment call, low risk):
+  a changed flowchart now augments rather than showing the whole-block badge.
+
+## Operator plate (loose ends to pick up)
+
+- Session **0024** still unfinalized (carried since session 0025).
+- Stray untracked `specs/coauthoring-out-of-workspace.md` in the **code** repo —
+  a duplicate of the F8 spec that already lives in the content repo.
+- F9 was never captured as an issue; #21 / #1 issue hygiene.
+- Manual webview-render smoke for F7.1 (`docs/MANUAL-SMOKE-F7.1.md`) not yet run
+  by a human (colors paint — not auto-testable in the sealed sandbox).
+
+## Next /goal
+
+No open issues and no roadmap-driven next feature remain. Suggested next:
+a **capture session** to re-stock the backlog from the deferred F7 follow-ups —
+more mermaid diagram types for #22 (class/state/ER/gantt), preview→source
+scroll-sync, non-markdown rendered views — and to clear the hygiene loose ends
+above. Or pick one deferred item directly to plan-and-execute.
@@ -0,0 +1,102 @@
+# Session 0028.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-11T21-25 (PST)
+> End: 2026-06-11T21-31 (PST)
+> Type: capture (tracked-lite)
+> Posture: careful (capture default, §6.5)
+> Status: **FINALIZED**
+
+## Launch prompt
+
+```
+Capture session: file an issue for the markdown cowriting product vision —
+clean zero-annotation editor (left) + interactive annotated markdown preview
+(right) as the single review surface; annotations on/off toggle; per-author
+track-changes (green=human, blue=LLM, strikethrough=deleted) with ✓/✗
+accept-reject on LLM changes only, driven from inside the preview. Markdown-only.
+```
+
+## Outcome (one line)
+
+Filed **issue #29** (`type/feature`, `priority/P1`) — "Interactive track-changes
+review in the markdown preview (clean editor + accept/reject)" — capturing a
+product-defining vision developed in a brainstorm during session 0027's
+conversation.
+
+## Pre-state
+
+- Came directly out of session 0027 (which finalized just before): the operator
+  had hit the F6 two-pane diff (`ctrl+option+d`) with F3 attribution decorations
+  bleeding into it and said "I don't understand this."
+- That sparked a brainstorm (in 0027's conversation, before this capture session
+  was claimed): single-pane Word-style track changes → in-editor inline → and
+  finally the clean-editor / interactive-annotated-preview split, markdown-only.
+- No open issues on the tracker (Epic #1 + F2–F9 + #22 all shipped).
+
+## Arc (turn by turn)
+
+1. **Diagnosis (carried from 0027).** Identified the confusing screenshot as two
+   plugin layers colliding: F6 diff (red/green) + F3 attribution (blue Claude fill
+   / green human border) on the same text. Confirmed the blue from the code
+   (`AGENT_DECO = rgba(99,102,241,.18)`).
+2. **Brainstorm (superpowers:brainstorming).** Walked the operator from "single
+   pane, strikethrough, Word-like" through the forks:
+   - Surfaced that F7 (`ctrl+option+r`) already is a single-pane rendered
+     track-changes view; F9 already adds an author-coloring toggle.
+   - Operator wanted **B** (in-editor inline) + **accept/reject of LLM changes
+     only** → mapped onto F3 (who) + F4 (accept/reject).
+   - Operator disliked F3's editor border → led to "color only the delta, by
+     author" and then the realization that the **editor should carry zero
+     annotation** and all review lives in the **preview**.
+   - Operator locked **markdown-only** ("this should really be the
+     vscode-markdown-cowriting-plugin") — F6 deprecated as a view; **rename
+     deferred**.
+   - Chose to **capture** (not spec now).
+3. **Capture (this session, wgl-capture).** Claimed tracked-lite session 0028;
+   resolved the content repo; sized the ask as a **Feature** (needs design, one
+   coherent increment); drafted `issues/interactive-markdown-review-preview.md`
+   section by section; handed it for review.
+4. **File (on approval).** Operator said "file it." Fixed two script-config snags
+   (default host was `git.wiggleverse.org` → `--host git.benstull.org`; forced
+   token service resolved the wrong host's token → set
+   `WGL_CAPTURE_TOKEN_SERVICE=wgl-gitea-issues-readwrite-token-git.benstull.org`).
+   Ensured labels, filed **#29**, verified labels (`type/feature` + `priority/P1`).
+
+## Cut state
+
+- **Issue #29** open on `benstull/vscode-cowriting-plugin`:
+  https://git.benstull.org/benstull/vscode-cowriting-plugin/issues/29
+- Draft `issues/interactive-markdown-review-preview.md` stays **uncommitted** in
+  the content repo working tree (INV-8 — the Author's to publish/discard).
+- Code repo `main` clean & pushed (only the carried-in stray `specs/` untracked).
+- No spec/plan artifact submitted (INV-6, tracked-lite).
+
+## Tooling notes (for the next capture on a benstull-hosted app)
+
+`ensure-capture-labels.sh` / `capture-issues.sh` default to `git.wiggleverse.org`
+and to the host-less issue token. For benstull apps, pass `--host git.benstull.org`
+**and** `WGL_CAPTURE_TOKEN_SERVICE=wgl-gitea-issues-readwrite-token-git.benstull.org`.
+(Candidate fix worth a plugin-feedback issue: auto-derive the host from the
+resolved app remote.)
+
+## Deferred decisions
+
+_Careful posture — the operator approved the filing (INV-1) and the type (Feature)
+explicitly; nothing was auto-decided that needs review._
+
+## Operator plate (loose ends)
+
+- Session **0024** still unfinalized.
+- Pre-existing uncommitted content-repo state not from this session:
+  `M specs/coauthoring-diff-view.md` + several untracked `issues/*.md` drafts.
+- Stray `specs/coauthoring-out-of-workspace.md` in the **code** repo working tree.
+- Capture-script host/token defaults (above) — candidate plugin-feedback.
+
+## Next /goal
+
+**Brainstorm / spec #29.** It's a `type/feature`, so it needs a **Solution Design
+before plan-and-execute** (R3) — and most of the design already exists in the
+issue body + memory. (Operator has in fact asked to brainstorm #29 next.) The
+rename to `vscode-markdown-cowriting-plugin` is a separate deliberate gesture if
+wanted.
@@ -0,0 +1,113 @@
+# Session 0029.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-11T21-32 (PST)
+> End: 2026-06-11T21-53 (PST)
+> Type: brainstorming
+> Poster: Ben Stull (with Claude)
+> Posture: autonomous (yolo)
+> Status: **FINALIZED**
+
+## Launch prompt
+
+```
+/goal next   →  (resolved from memory) Brainstorm + write the Solution Design for
+Feature #29 — interactive track-changes review in the markdown preview: clean
+zero-annotation editor (left) + the rendered markdown preview (right) as the single
+review surface; annotations on/off toggle; changes-since-baseline painted
+green=human / blue=LLM / strikethrough=deleted; ✓/✗ accept-reject of the LLM's
+changes ONLY, driven from inside the preview webview. Markdown-only product (F6
+deprecated as a view; rename deferred). Mostly assembly of F3/F4/F6/F7/F9 + 3 new
+pieces. Per issue #29.
+```
+
+## Pre-state
+
+- Epic #1 shipped (F1–F9 + F7.1). Last work: F7.1 intra-diagram mermaid diff (#22,
+  session 0027), then F9 authorship preview (#27, session 0026).
+- #29 (F10) had been **captured** as a `type/feature` (P1) in session 0028, with a
+  rich issue body + a draft in the content repo (`issues/interactive-markdown-review-preview.md`).
+- A brainstorming placeholder **0029** had already been claimed at 21:32 (just
+  before a `/clear`), then the conversation was cleared and reopened with
+  `/goal next`. Clean `main`, 0 ahead/0 behind origin.
+
+## Session arc (turn-by-turn)
+
+1. **Route.** `/goal next` resolved the memory `Next /goal:` → "brainstorm/spec
+   #29". A Feature needs a Solution Design first (R3) → **brainstorming** session.
+   Invoked `wgl-brainstorming`.
+2. **Adopt, don't duplicate.** The claim dry-run surfaced an in-flight 0029
+   placeholder (`Type: brainstorming`, same task, claimed 4 min earlier — the
+   orphan from the pre-`/clear` attempt). Decision: **adopt 0029** rather than
+   claim a redundant 0030. Verified clean baseline; CLAUDE.md import present;
+   resolved app (content repo present, no BDD corpus).
+3. **Orient.** Read issue #29's content-repo draft, the F7 rendered-preview spec
+   (`coauthoring-rendered-preview.md`, incl. §11 F7.1), the F9 authorship spec
+   (`2026-06-11-authorship-preview-design.md`), and dispatched an Explore agent
+   that mapped every code seam (F3 decorations + `spansFor`; F7
+   `renderTrackChanges`/`renderAuthorship`/`renderOp`; F4 `accept`/`reject` →
+   `applyAgentEdit`; F6 baseline + `onDidChangeBaseline`; webview
+   `postMessage`/`setMode`; commands/keybindings; tests).
+4. **Found the crux.** F9 had *explicitly rejected* combining authorship with the
+   diff (INV-26: "two modes, never combined"); #29 *reverses* that. And the issue
+   is internally inconsistent: F4 ships **propose-by-default** (LLM edits stay
+   pending, not in the buffer), yet the issue says blue = "changes since the
+   baseline" and "✗ reverts the document" — wording that imagines LLM text already
+   in the buffer.
+5. **Resolved the forks with the operator** (`superpowers:brainstorming` process):
+   - **Linchpin** — *where do the LLM's ✓/✗ changes come from?* → **Proposals
+     surface in the preview** (keep propose-by-default; ✓ = F4 accept + baseline
+     advance, ✗ = F4 reject). [vs. apply-then-review / both-axes.]
+   - **Clean-editor scope** → **fully clean** (strip F3 attribution *and* F4
+     proposal threads; proposals become preview-only).
+   - **F6 disposition** → **hide** command + keybinding (`when:false`); keep
+     controller + baseline data layer.
+6. **Two follow-ons decided autonomously** (logged): status-bar indicator (not
+   forced auto-open) for proposal-while-no-preview; remove F9's dead
+   `renderAuthorship`/segmented control (salvage the PUA-sentinel coloring into a
+   shared `colorByAuthor`); neutral strikethrough for deletions.
+7. **Presented the design**, then (autonomous) wrote the Solution Design:
+   `specs/coauthoring-interactive-review.md` (v0.1.0) — architecture (one pure
+   `renderReview` overlaying diff + author sentinels + proposal blocks; interactive
+   sealed webview posting intent to the F4 seam), INV-32/33/34, 4 slices, host-E2E
+   tier. Self-reviewed (no placeholders; INV numbering continues from F7.1's 29–31
+   / F9's 26–28; issue-wording reconciliation stated explicitly).
+8. **Finalized** (this skill): submitted the spec graduated to the content repo;
+   updated memory; published this transcript.
+
+## Cut state
+
+- **Deliverable:** `specs/coauthoring-interactive-review.md` (F10, v0.1.0) →
+  submitted **graduated** to `vscode-cowriting-plugin-content/specs/` (commit
+  `118bc29`).
+- **Code:** none (brainstorming session — no implementation).
+- **Memory:** new `f10-interactive-review-spec-graduated.md` + index + `Next /goal:`.
+- Working tree: clean `main` apart from this transcript and the untracked `specs/`
+  scratch (the submitted spec + the pre-existing stray F8 spec).
+
+## Deferred decisions
+
+_Autonomous-mode low-confidence calls the driver made and would have
+liked operator input on._
+
+- **Proposal-while-no-preview affordance → status-bar indicator** (clickable to
+  open), rather than forcing the preview to auto-open. Respects "write left, review
+  right"; cheap to flip to auto-open. (Spec §6.7, §9.)
+- **F9's now-dead `renderAuthorship` + segmented control → removed** (not left
+  dormant), salvaging the PUA-sentinel author-coloring into a shared
+  `colorByAuthor` helper `renderReview` reuses. (Spec §6.2, §9.)
+- **Deletions → neutral strikethrough** (not author-colored), per the issue's
+  "strikethrough = deleted". (Spec §6.7.)
+
+## Operator plate (loose ends to be aware of)
+
+- **Session 0024 unfinalized** — still open from F8 work; not touched here.
+- **Stray `specs/coauthoring-out-of-workspace.md`** (already-graduated F8 spec)
+  sits untracked in the CODE repo; left untouched (submit-spec took only the F10
+  file). A deliberate cleanup PR can remove it.
+
+## Next-session prompt
+
+```
+/goal plan-and-execute #29 (F10 interactive review) per vscode-cowriting-plugin-content/specs/coauthoring-interactive-review.md — write the implementation plan from §7 (SLICE-1 clean editor → SLICE-2 renderReview engine → SLICE-3 interactive controller+webview → SLICE-4 tests/docs) and execute; host-E2E tier, no deploy pipeline (extension, no flotilla)
+```
@@ -0,0 +1,141 @@
+# Session 0030.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-11T23-48 (PST)
+> End: 2026-06-12T00-42 (PST)
+> Type: planning-and-executing
+> Posture: autonomous (yolo)
+> Status: **FINALIZED**
+> Outcome: **F10 (#29) shipped to `main`** via PR #30; issue #29 closed.
+
+## Launch prompt
+
+```
+/goal next → plan-and-execute #29 (F10 — interactive markdown review).
+Spec: specs/coauthoring-interactive-review.md (graduated in brainstorming session 0029).
+Markdown preview as the single interactive review surface: clean zero-annotation
+editor + annotated preview with on/off toggle (green=human / blue=LLM / strikethrough),
+✓/✗ accept-reject for LLM-only changes in the webview. LLM changes surfaced as pending
+F4 proposals (✓=F4 accept, ✗=F4 reject). One pure renderReview overlay; reverses F9
+INV-26; INV-32..34; 4 slices.
+```
+
+`/goal next` resolved the `Next /goal:` field recorded by session 0029's finalize
+to "plan-and-execute #29 (F10)".
+
+## Pre-state
+
+- On `main`, clean (origin/main behind 1 — fast-forwarded). Anchor #29 is
+  `type/feature` with a **graduated** Solution Design
+  (`specs/coauthoring-interactive-review.md`) → R2/R3 gate satisfied.
+- F2–F9 + F7.1 shipped; F10 is "mostly assembly" of F3 (attribution `spansFor`),
+  F4 (propose/accept seam), F6 (baseline), F7/F9 (render engine) + 3 new pieces:
+  strip editor decorations, collapse F9's two modes to one on/off toggle, make the
+  preview interactive.
+
+## Plan
+
+Goal: plan-and-execute F10 (#29). "Write left, review right" — clean editor, the
+rendered preview as the single interactive review surface. 4 slices (spec §7.2),
+14 tasks; plan written via `superpowers:writing-plans`, archived to the content
+repo `plans/` at finalize. Executed via `superpowers:subagent-driven-development`
+(fresh implementer per slice + spec-compliance and code-quality review each).
+
+**Code reconciliations (spec vs. actual), discovered by reading the code first:**
+public seams are `acceptById(docPath,id)`/`rejectById(docPath,id)` (not private
+`accept`/`reject`); `Proposal` carries `anchorId` resolved via `artifact.anchors`
+→ `resolve(text, fp)` (not literal offsets); no `onDidChangeProposals` existed (added);
+`trackChangesPreview` already took `AttributionController` (F9), added `ProposalController`;
+`extension.ts` constructed the preview BEFORE the proposal controller — reordered.
+
+## Session arc (turn-by-turn)
+
+1. **Init.** `/goal next` → routed to `wgl-planning-and-executing`. Claimed session
+   0030 (peek: nothing in flight). Fast-forwarded `main`. Read the F10 spec + mapped
+   the 8 files F10 touches via 3 parallel Explore agents (exact signatures), then
+   read the central bodies (`trackChangesModel.ts`, `trackChangesPreview.ts`,
+   `proposalController.ts`, `extension.ts`) for ground truth.
+2. **Plan.** Wrote `docs/superpowers/plans/2026-06-11-f10-interactive-review.md`
+   (4 slices, 14 tasks, TDD on the pure engine, host-E2E for the vscode layer).
+   Branch `f10-interactive-review`; committed the plan.
+3. **SLICE-1 (clean editor, INV-32).** Implementer stripped F3 attribution
+   decorations (kept `spansFor`), removed F4 in-editor comment threads + pending
+   decoration (refactored bookkeeping to `state.live`/`state.unresolved`), hid the
+   F6 diff command/`ctrl+alt+d` + attribution toggle. Spec review ✅. **Code-quality
+   review found** the `visible`/`toggle`/`isVisible`/`toggleAttribution` toggle was
+   now dead/misleading → fixed directly (retired it; spec §6.2 says so). 189 unit green.
+4. **SLICE-2 (combined render engine, INV-33).** Strict TDD. Extracted
+   `colorByAuthor` from the F9 sentinels; added `renderPlain` + `renderReview`
+   (+ `ProposalView`). Deferred removing public `renderAuthorship` to SLICE-3 (kept
+   the slice green). Spec ✅. **Code-quality review found a real correctness bug:**
+   `renderReview` matched a block's source range by raw-string equality →
+   misaligned author coloring for duplicate paragraphs. Fixed (TDD, positional
+   range pointer) + collapsed a redundant branch + escaped the proposal id. 41 model
+   / 198 total green.
+5. **SLICE-3 (interactive controller + webview, INV-34).** Removed public
+   `renderAuthorship`; added `ProposalController.listProposals`/`onDidChangeProposals`/
+   `keyFor`; collapsed the preview to an on/off toggle taking `ProposalController`;
+   routed ✓/✗/setMode webview messages through `acceptById`/`rejectById` (webview
+   never mutates the doc); status-bar indicator (PUC-6); rebuilt the webview asset
+   (on/off checkbox + ✓/✗ click→postMessage + CSS); reordered `extension.ts`. Spec ✅
+   (INV-34 traced end-to-end). Code-quality ✅ with one polish: renamed the webview
+   panel title "Track changes" → "Review" (fixed directly).
+6. **SLICE-4 (tests & docs).** Updated obsolete suites (F9 authorship-mode +
+   retired-toggle), added `test/e2e/suite/f10Review.test.ts`, wrote
+   `docs/MANUAL-SMOKE-F10.md` + README F10 section. Found + fixed one real seam bug:
+   `statusText()` returned stale text after `hide()` → added `hideStatus()` clearing
+   the text. **E2E green: 46 + 5 passing.** Spec ✅, code-quality ✅.
+7. **Final review + verification.** Whole-implementation review = ready to merge
+   (all invariants verified across slice seams). Ran the full suite myself:
+   189 unit + 51 E2E + clean build, tree clean.
+8. **Ship.** Pushed branch; created PR #30 (Gitea API); merged to `main` (HTTP 200);
+   synced `main`, deleted the branch (local + remote); commented + closed #29;
+   post-merge sanity (build + 189 unit) green.
+9. **Finalize.** Archived the plan to the content repo `plans/`; updated memory;
+   published this transcript.
+
+## Cut state (final)
+
+- **`main`** carries F10 (merge commit `4160890`, PR #30). Working tree clean
+  except the pre-existing untracked `specs/` dir.
+- **Tests:** 189 unit (vitest, vscode-free) + 51 host E2E + clean esbuild build.
+- **Tracker:** issue #29 closed; **no open issues remain** (Epic #1 closed).
+- **No deploy pipeline** — VS Code extension, no flotilla/PPE/prod stage; "done" =
+  merge to `main` per spec §7.3. Manual webview smoke (`docs/MANUAL-SMOKE-F10.md`)
+  is the only step not automatable (sealed sandbox) — left for the operator.
+
+## Deferred decisions
+
+- **Proposal placement: trailing blocks vs. inline-at-anchor.** Spec §2/§6.2
+  describes pending proposals "rendered **inline at its resolved anchor**." The
+  implemented `renderReview` instead appends ALL proposals as trailing
+  `cw-proposal` blocks (anchored before unanchored), per the plan's explicit
+  Task-6 v1 design note (determinism + simplicity; inline-injection into the
+  block-level diff body is materially more complex). INV-34 holds (proposals are
+  visible, carry ✓/✗, never dropped); the deviation is purely positional. The
+  final whole-impl review judged it defensible and non-blocking. **Deferred:**
+  inline-at-anchor placement of resolved proposals → candidate follow-up issue,
+  alongside the spec §9 open items (preview→source scroll-sync, intra-emphasis
+  sentinel hardening, deleting F6's dead two-pane view code, the repo rename).
+  Worth confirming with the operator whether to file these.
+
+## Loose ends
+
+- Stray untracked `specs/` dir in the code repo (F8 + F10 design working copies;
+  they belong in the content repo `specs/` collection and were already submitted
+  there) — left untouched (carried from sessions 0024/0027).
+- Stray worktree `.claude/worktrees/pedantic-brahmagupta-472309/` from a prior
+  session — not cleaned up here.
+- Session 0024 still unfinalized (carried).
+
+## Next-session prompt
+
+The F10 follow-ups are not filed as issues and the tracker has no open anchor, so
+the next move is a **capture** session to file them (issue-anchored discipline, R2):
+
+```
+/goal capture the F10 follow-ups as Gitea issues — lead: inline-at-anchor placement of resolved proposals in the review preview (per specs/coauthoring-interactive-review.md §2/§6.2; today they render as trailing cw-proposal blocks); plus the §9 open items — preview→source scroll-sync, intra-emphasis sentinel hardening, deleting F6's dead two-pane view code, and the repo rename to vscode-markdown-cowriting-plugin
+```
+
+Read [[f10-interactive-review-shipped]] first. Then a follow-up
+planning-and-executing session can take whichever filed issue the operator picks.
@@ -0,0 +1,101 @@
+# Session 0031.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-12T01-06 (PST)
+> End: 2026-06-12T01-10 (PST)
+> Type: capture (tracked-lite)
+> Posture: autonomous (yolo) — driven by an explicit `/goal` (launch-prompt override of the capture careful-default)
+> Status: **FINALIZED**
+> Outcome: **5 F10 follow-up issues filed** to the tracker — #31–#35.
+
+## Launch prompt
+
+```
+/goal capture the F10 follow-ups as Gitea issues (from session 0030):
+(1) inline-at-anchor placement of resolved proposals in the review preview (lead; spec §2/§6.2 — today trailing cw-proposal blocks);
+(2) preview→source scroll-sync (spec §9);
+(3) intra-emphasis sentinel hardening (F9 deferred refinement);
+(4) delete F6's dead two-pane view code (hidden when:false in F10);
+(5) repo rename to vscode-markdown-cowriting-plugin (deferred, spec §1.7).
+Tracker: benstull/vscode-cowriting-plugin.
+```
+
+The goal carried over from session 0030's finalize as the active `/goal`; a Stop
+hook held the turn until the issues were actually filed (the prior session had only
+*identified and deferred* them).
+
+## Pre-state
+
+- Immediately follows planning-and-executing session 0030 (F10 / #29 shipped to
+  `main`). The tracker had **no open issues** (Epic #1 closed); the F10 follow-ups
+  existed only as deferred decisions in 0030's transcript + memory.
+- Content repo `vscode-cowriting-plugin-content` cloned; `issues/` collection holds
+  prior capture drafts (the template reference).
+
+## What happened
+
+1. **Claim.** `/goal` → classified as **capture** → `wgl-capture`. Claimed
+   tracked-lite session 0031 (nothing else in flight). Resolved the content repo.
+2. **Context (no external signal).** Worked from 0030's deferred decisions + the
+   F10 spec (`coauthoring-interactive-review.md` §2/§6.2/§9/§1.7) — self-originated
+   capture (BUC-3), origin = this session + session 0030.
+3. **Size-challenge.** Five discrete asks, typed by the RFC §4.1 sizing test:
+   - #31 inline-at-anchor proposal placement → **type/feature** P2 (user-facing
+     review-UX fidelity; the lead).
+   - #32 preview→source scroll-sync → **type/feature** P3.
+   - #33 intra-emphasis sentinel hardening → **type/task** P3 (robustness, no
+     independent business value).
+   - #34 delete F6's dead two-pane view code → **type/task** P3 (refactor/cleanup).
+   - #35 repo rename to `vscode-markdown-cowriting-plugin` → **type/task** P3
+     (coordinated rename chore).
+4. **Draft + file.** Authored each draft (§5 template) into the content repo
+   `issues/` working tree, then filed via `capture-issues.sh` (labels ensured
+   idempotently). Each issue carries exactly one `type/*` + one `priority/*` and a
+   populated Source/signal (INV-2).
+5. **Filed:** #31 (feature/P2), #32 (feature/P3), #33 (task/P3), #34 (task/P3),
+   #35 (task/P3) — all on `benstull/vscode-cowriting-plugin`.
+
+## Cut state (final)
+
+- **Issues #31–#35 open** on the tracker, correctly labelled.
+- **Draft files** (`issues/{f10-inline-anchor-proposals, f10-preview-scroll-sync,
+  intra-emphasis-sentinel-hardening, delete-f6-two-pane-view-code,
+  repo-rename-markdown-cowriting}.md`) remain **uncommitted** in the content-repo
+  working tree — the Author's to commit/publish or discard (INV-8). This skill did
+  not push them.
+- Code repo `main` clean (no code change this session). No spec/plan artifact
+  submitted (INV-6, tracked-lite).
+
+## Deferred decisions
+
+- **Autonomous type/priority assignment.** The capture default is *careful*
+  (operator-reviewed, draft-by-draft). Under the explicit `/goal` launch-prompt
+  override I drove this autonomously: I assigned each issue's `type/*` and
+  `priority/*` and filed without pausing for per-section operator review. The
+  operator may want to **re-triage** — especially #31's `type/feature` P2 (could be
+  a `type/task` if treated as pure render polish) and the relative priority order
+  of #32–#35. Issues are editable/closable, so this is low-risk.
+- **Issue-scoped token 401.** The dedicated capture token service
+  (`wgl-gitea-issues-readwrite-token`) returned HTTP 401 (expired or missing the
+  issue scope). Filed via the working admin token by setting
+  `WGL_CAPTURE_TOKEN_SERVICE=wgl-gitea-token-git.benstull.org`. The dedicated
+  issues token should be refreshed so future capture sessions don't need the
+  admin-token fallback.
+
+## Loose ends
+
+- Five `issues/*.md` drafts uncommitted in the content repo (Author to publish/discard).
+- Carried: stray untracked `specs/` dir + stray worktree
+  `.claude/worktrees/pedantic-brahmagupta-472309/` in the code repo; session 0024
+  still unfinalized.
+
+## Next-session prompt
+
+```
+/goal plan-and-execute #31 — render F10 review proposals inline at their resolved anchor (not trailing cw-proposal blocks), per specs/coauthoring-interactive-review.md §2/§6.2; renderReview proposal-block placement only, keep the unanchored fallback (INV-34) and determinism (INV-33)
+```
+
+Read [[f10-interactive-review-shipped]] first. #31 is a `type/feature` leaf (ready
+R2 anchor); the change is bounded to `renderReview`'s proposal-block placement in
+`src/trackChangesModel.ts`. Alternatively, triage #31–#35 priorities first if the
+operator wants to re-order before building.
@@ -0,0 +1,91 @@
+# Session 0032.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-12T01-20 (PST)
+> End: 2026-06-12T03-05 (PST)
+> Type: planning-and-executing
+> Posture: autonomous (yolo)
+> Status: **FINALIZED**
+
+## Launch prompt
+
+"What's next?" → operator chose: **tidy loose ends, then plan-and-execute #31**.
+Then "keep going" → continued into #34, then #33.
+
+## Pre-state
+
+On `main`, but the baseline was not clean: local `main` behind `origin/main` by 3
+(session-transcript commits — this app keeps `sessions/` in the code repo), an
+untracked `specs/` dir, an abandoned worktree `pedantic-brahmagupta-472309`
+(106 commits behind), and two merged-but-undeleted remote branches. Memory pointed
+to plan-and-execute #31 next.
+
+## Arc
+
+### 1. Loose-ends tidy
+- Fast-forwarded `main`; synced the 0032 claim.
+- F10 spec `coauthoring-interactive-review.md` was missing from the content repo —
+  placed it in the content repo **working tree** (left uncommitted; that repo has
+  operator-owned capture drafts + a diff-view.md edit pending the operator's publish).
+- Removed the stray code-repo `specs/` (one dup of content, one now-placed).
+- Removed the abandoned worktree (untracked `prototype/` + `.claude/` snapshotted
+  to `~/.wgl/abandoned-worktree-snapshots/pedantic-brahmagupta-472309/`).
+- Deleted the two merged remote branches.
+- **Correction:** the "0024 unfinalized" loose end carried in prior memory is
+  FALSE — 0024 has a finalized transcript. Dropped from memory.
+
+### 2. #31 — inline-at-anchor proposals (PR #36, merged; issue closed)
+TDD. `renderReview` now associates each resolved proposal with the current-side
+block its `anchorStart` falls in and emits the `cw-proposal` block right after that
+block; same-block order by anchorStart then id; unresolved/unplaceable still trail
+(INV-33/34). Implements the placement the F10 spec §2/§6.2 already specified (the
+#29 trailing-block was a recorded v1 deferral). +5 unit tests, +1 E2E in-place
+assertion. One pre-existing F7 E2E flake (fixed-timeout `settle()` race) passed on
+re-run — unrelated to the change (different code path).
+
+### 3. #34 — delete F6 dead two-pane view (PR #37, merged; issue closed)
+Removed the dead `vscode.diff` VIEW: `DiffViewController.toggle`/`findDiffTab`/
+`epochLabel`/`isDiffOpen`, the `cowriting-baseline:` content provider +
+`BASELINE_SCHEME` + `baselineUri` + content-provider emitter, the `toggleDiffView`
+command + `ctrl+alt+d` keybinding. Kept the baseline DATA layer fully intact
+(ensureBaseline/advance/pin/capture, getBaseline, onDidChangeBaseline, persistence
+INV-19, machine-landing advance INV-18). Rewrote the diffView E2E suite to the
+baseline-data-layer tests; flipped F10 + no-workspace assertions to "toggleDiffView
+absent."
+
+### 4. #33 — intra-emphasis sentinel hardening (characterized, DEFERRED)
+Reproduced two failure modes with the real renderer (recorded as issue #33
+comment). Operator chose to defer to its own session. No code changed.
+
+## Cut state
+
+`main` clean and synced; 0 open PRs. #31 + #34 shipped and closed. 194 unit + 49
+E2E green; typecheck + build clean. No deploy pipeline (VS Code extension, no
+flotilla/PPE).
+
+## Deferred decisions
+
+1. **Kept `pinDiffBaseline` in #34 against the issue's literal acceptance.**
+   Alternative: remove it as the issue listed. Why: the canonical Solution Design
+   (spec §6.7) scopes the removal to the two-pane VIEW only ("keep the controller +
+   baseline store"); `pin()` is a §6.4 baseline-lifecycle op, never touches
+   `vscode.diff`, and is exercised by live F7 baseline-reset tests. Spec wins over
+   the P3 capture draft (documentation-leads-automation). Logged for operator
+   awareness.
+2. **F10 spec left UNCOMMITTED in the content repo working tree** rather than
+   committed/pushed. Alternative: push it. Why: the content repo has operator-owned
+   pending work (capture issue-drafts + a diff-view.md edit); content publishing is
+   the Author's gesture (per wgl-capture). Surfaced to the operator.
+3. **Discarded the abandoned worktree's untracked files** (prototype/index.html +
+   .claude config). Snapshot-and-proceed: copied to
+   `~/.wgl/abandoned-worktree-snapshots/` before `git worktree remove --force`.
+
+## Next /goal
+
+```
+/goal plan-and-execute #33 (intra-emphasis sentinel hardening) via the token-aware HTML-post-process approach recommended in the issue #33 comment, per specs/coauthoring-interactive-review.md §1.7/§9
+```
+
+(Alternatives in the backlog: #32 scroll-sync needs a Solution Design first
+(brainstorming); #35 repo rename is an outward admin gesture awaiting operator
+timing.)
@@ -0,0 +1,59 @@
+# Session 0033.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-12T03-14 (PST)
+> End: 2026-06-12T03-27 (PST)
+> Type: capture (tracked-lite)
+> Posture: careful (capture default — output operator-reviewed)
+> Status: **FINALIZED**
+
+## Launch prompt
+
+`/wgl-capture` — "Undo doesn't render correctly in the preview pane".
+
+## Arc
+
+A capture session turning one operator observation into a filed, typed,
+source-linked Gitea issue.
+
+- **Clarified the symptom** (AskUserQuestion): "updates but wrong marks" in the
+  **F10 review preview**, reproduces **always** on undo. Reflected back the read
+  that undo reverts the buffer (text updates) but the overlay marks (since-baseline
+  diff / F3 author-coloring / F4 proposal blocks) go stale/wrong.
+- **Drafted** `issues/undo-renders-wrong-marks-in-preview.md` in the content repo
+  working tree (terse, bug-depth §5 template). Classified **type/bug**.
+- **Type/bug vs the tool:** `capture-issues.sh` only accepts
+  `epic|feature|story|task`, so it can't file a bug — even though the tracker has a
+  `type/bug` label and handbook §4.3's gate vocabulary includes "leaf
+  story/task/bug." Declined to mis-type the defect as a task to fit the tool.
+- **Filed `#38`** (type/bug, priority/P1 — operator decision) via `gitea-api.sh`
+  directly (labels API), since the deterministic capture script couldn't express
+  `bug`. → https://git.benstull.org/benstull/vscode-cowriting-plugin/issues/38
+- **Filed plugin feedback `#124`** (friction/medium) on the plugin tracker for the
+  capture-can't-type-bugs gap, framed as a taxonomy inconsistency to reconcile
+  across the Capture RFC §4, `ensure-capture-labels.sh`, `capture-issues.sh`,
+  `wgl-capture/SKILL.md`, and handbook §4.3.
+
+## Cut state
+
+Code repo clean on `main`. One issue filed (#38) + one plugin-feedback issue (#124).
+Draft `undo-renders-wrong-marks-in-preview.md` left uncommitted in the content repo
+working tree (operator's to publish — INV-8), alongside the other pending capture
+drafts + the F10 spec placed in session 0032.
+
+## Deferred decisions
+
+1. **Filed #38 with `type/bug` via a `gitea-api.sh` bypass** rather than
+   `capture-issues.sh`. Alternative: type it `task` to stay within the tool. Why:
+   the defect's honest type is `bug` (label exists; §4.3 recognizes it); the script
+   gap shouldn't drive mis-classification. The gap itself was logged as plugin
+   feedback #124. Priority P1 was the operator's explicit call.
+
+## Next /goal
+
+```
+/goal plan-and-execute #38 (undo renders wrong marks in the F10 review preview, P1) — reproduce the wrong-marks-after-undo case, find which layer (F3 attribution spans / F4 proposal anchors) goes stale on undo and reconcile it, confirm across edit→undo→redo
+```
+
+(P1 — supersedes session 0032's #33 pointer. #33 intra-emphasis hardening (P3) and
+#32 scroll-sync (feature, needs a Solution Design) remain open.)
@@ -0,0 +1,63 @@
+# Session 0034.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-12T03-28 (PST)
+> End: 2026-06-12T03-39 (PST)
+> Type: planning-and-executing
+> Posture: autonomous (yolo)
+> Status: **FINALIZED**
+
+## Launch prompt
+
+`/goal next` → resolved from memory to: plan-and-execute #38 (undo renders wrong
+marks in the F10 review preview, P1).
+
+## Plan
+
+Fix #38 (P1). Root cause (confirmed by reading + E2E repro): attribution attributed
+every non-seam change to currentAuthor() (always human), ignoring e.reason — so
+undo/redo re-inserting text falsely colored it human. Fix: geometry-only span
+reconciliation on undo/redo (no fresh attribution). E2E repro → minimal fix → green.
+
+## Arc (systematic debugging)
+
+- **Phase 1 (root cause):** traced the three data sources feeding the pure
+  `renderReview` (baseline diff / F3 author spans / F4 proposals). Read
+  `attributionController.onDidChange` + `attributionTracker.applyChange`: every
+  non-seam change is attributed to `currentAuthor()` (always human) and `e.reason`
+  is ignored. Proposals only shift (not the culprit). Disk-sync guard masks
+  undo-to-saved-state → bug is on mid-edit undo.
+- **Reproduce:** wrote `test/e2e/suite/undoMarks.test.ts` driving edit→edit→undo
+  (mid-edit, buffer stays dirty). Confirmed FAILING: after undo, `spansFor` =
+  `[{start:6,end:12,author:"human"}]` over restored "bravo " — false attribution.
+- **Fix (Phase 4):** `applyChange` gains `attributeInserted` (default true);
+  `onDidChange` sets it false and skips seam matching when
+  `e.reason === Undo|Redo`. Restored text stays neutral, not falsely human.
+- **Verify:** repro passes; +3 unit tests for the geometric-only path. 197 unit +
+  50 E2E green; typecheck clean.
+- **Shipped:** PR #39 merged → main; issue #38 closed.
+- **Follow-up filed #40** (type/task, P3): restore *exact* prior provenance on
+  undo/redo (the fix leaves undone text neutral, not its original author color);
+  needs an attribution history stack synced to the editor undo stack.
+
+## Cut state
+
+`main` clean + synced; 0 open PRs. #38 fixed and closed; #40 filed. 197 unit + 50
+E2E green. No deploy pipeline (VS Code extension).
+
+## Deferred decisions
+
+1. **Scoped the fix to "neutral on undo," not exact provenance restoration.**
+   Alternative: an attribution history stack that restores each char's original
+   author on undo/redo. Why deferred: synchronizing such a stack with VS Code's
+   undo (which coalesces edits) is fragile, and the neutral fix already removes the
+   reported misleading marks. Tracked as #40 (P3) so it isn't lost.
+
+## Next /goal
+
+```
+/goal plan-and-execute #33 (intra-emphasis sentinel hardening) via the token-aware HTML-post-process approach recommended in the issue #33 comment, per specs/coauthoring-interactive-review.md §1.7/§9
+```
+
+(Backlog: #40 undo provenance (P3); #32 scroll-sync (feature, needs a Solution
+Design); #35 repo rename (outward, operator timing).)
@@ -0,0 +1,104 @@
+# Session 0035.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-12T08-53 (PST)
+> End: 2026-06-12T10-58 (PST)
+> Type: capture (tracked-lite)
+> Status: **FINALIZED**
+
+## Launch prompt
+
+```
+Add "Open Cowriting Review Panel" to the context window when you right click on a markdown file or its tab
+```
+
+## Pre-state
+
+- Branch `main`, clean, pushed. Prior session 0034 had shipped the #38 undo fix
+  (PR #39). Recorded next goal there was #33 (intra-emphasis hardening) — this
+  session instead opened as a **capture** session on a new line of UX asks.
+- Existing command surface (from `package.json`): `cowriting.editSelection`
+  ("Ask Claude to Edit Selection", editor-context, selection-gated);
+  `cowriting.showTrackChangesPreview` ("Open Review Preview", palette +
+  `ctrl+alt+r`); `cowriting.pinDiffBaseline` ("Pin Review Baseline to Now"),
+  registered but **palette-suppressed** (`when:false`) with no menu/keybinding —
+  orphaned since #34 deleted its F6 two-pane host.
+
+## Arc (turn by turn)
+
+1. **Claim + resolve.** Claimed tracked-lite transcript 0035 (`--type capture`).
+   Resolved app `vscode-cowriting-plugin`, content repo
+   `benstull/vscode-cowriting-plugin-content` (cloned), tracker
+   `benstull/vscode-cowriting-plugin`.
+2. **Ask 1 → #41.** Right-click on a markdown file/tab → "Open Cowriting Review
+   Panel". Sized **type/story, P2** (pure menu wiring to the existing preview
+   command). Drafted `issues/review-panel-context-menu.md`. Flagged the label
+   wrinkle (VS Code renders the command's own title). Operator: "first one looks
+   good" → ensured labels (needed host token
+   `wgl-gitea-issues-readwrite-token-git.benstull.org`; bare default 401'd) →
+   filed **#41**.
+3. **Ask 2 → #42.** "Ask Claude to Edit Document" when right-clicking the body with
+   no selection; Edit Selection/Edit Document on the tab too. Sized
+   **type/feature, P2** — beyond menu wiring it introduces a *new whole-document
+   edit capability* with open product questions (instruction gathering, one big
+   F4 proposal vs many, large-doc/token behavior). Drafted
+   `issues/ask-claude-edit-document-context-menu.md`. Operator: "file as a
+   feature" → filed **#42**.
+4. **Question: pin baseline.** Operator asked if the baseline can be reset to the
+   current editor. Answer: yes — `cowriting.pinDiffBaseline` does exactly that
+   (`capture(doc,"pinned")` → baseline := `document.getText()`). Initially said
+   "palette-only"; operator's screenshot showed it absent from the palette →
+   re-checked and **corrected**: it's `when:false`-suppressed with no menu/
+   keybinding, i.e. **unreachable from any UI**.
+5. **Ask 3 → #43 (lead).** Operator's direction: make the **rendered preview the
+   primary interaction surface** — toolbar (home of the annotations checkbox)
+   gains one **adaptive "Ask Claude…" button** (Edit Selection ⇆ Edit Document by
+   *preview* selection) and a **Pin baseline** button. Asked two shaping
+   questions; operator chose **coexist** (native menus #41/#42 become **gateways
+   that open the preview**) and **P1**. Sized **type/feature, P1**; central design
+   risk = mapping a rendered-preview selection back to a source markdown range
+   (fallback: document-level button first). Drafted
+   `issues/preview-toolbar-interaction-surface.md` → operator "file as-is" →
+   filed **#43**.
+6. **Dependencies.** Operator: make #41/#42 dependent on #43. Set via Gitea
+   issue-dependencies API — discovered the body needs `{"owner","repo","index"}`
+   with the repo key **`repo`** (documented `name` 404s as
+   `IsErrRepoNotExist`). #41 and #42 now **blocked-by #43**; verified.
+7. **Finalize** (tracked-lite): base survey (session repo clean, nothing to land);
+   no spec/plan artifact (INV-6); memory updated; transcript published.
+
+## Cut state
+
+- **Filed:** #41 (story P2), #42 (feature P2), #43 (feature P1) on
+  `benstull/vscode-cowriting-plugin`. #41 & #42 **blocked-by #43**.
+- **Session repo:** `main`, clean, unchanged (no code touched this session).
+- **Content repo:** 3 new `issues/*.md` drafts left **uncommitted** (Author's to
+  publish/discard — INV-8). No commits/pushes made to it.
+
+## Operator plate (loose ends)
+
+- The pin-baseline UI-reachability gap is captured only *inside* #43's scope
+  (operator chose to fold it in, not file a standalone bug).
+- #41/#42 carry no comment cross-linking #43; the dependency link is recorded in
+  Gitea instead. (Offered to add comments; not requested.)
+- Pre-existing untracked drafts from earlier capture sessions still sit in the
+  content repo working tree (not this session's concern).
+
+## Deferred decisions
+
+_None — all sizing/priority/relationship calls were made with the operator in the
+moment (sizing of #42 as feature, #43 coexist-vs-supersede and P1, folding the
+pin gap into #43)._
+
+## Next-session prompt
+
+```
+/wgl-brainstorming design #43 — preview toolbar as the primary interaction surface
+(adaptive "Ask Claude…" button: Edit Selection ⇆ Edit Document by preview-selection
+state; Pin baseline button; annotations checkbox stays). Central design problem:
+map a selection in the rendered (markdown-it) preview back to a source markdown
+range; fallback is ship document-level Ask-Claude + Pin first, selection-aware
+second. Coexists with #41/#42 (native right-click entries become gateways that
+open the preview). Builds on F7/F10 preview + annotations toggle, F3/F4 inner
+loop, F6 baseline store (gives orphaned cowriting.pinDiffBaseline a UI home).
+```
@@ -0,0 +1,105 @@
+# Session 0036.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-12T11-01 (PST)
+> End: 2026-06-12T12-18 (PST)
+> Type: brainstorming
+> Status: **FINALIZED**
+
+## Launch prompt
+
+`/goal next` — resume the next recorded goal. Memory (session 0035) recorded
+**Next /goal = design #43**: write the Solution Design for Feature #43 (F11) —
+the preview toolbar as the primary interaction surface (adaptive Ask-Claude +
+Pin-baseline buttons).
+
+## Summary
+
+A brainstorming session that produced and graduated the **F11 Solution Design**
+for Feature `#43`: making the F10 rendered-review preview's header toolbar the
+**primary interaction surface**. Beside the existing annotations on/off checkbox,
+the toolbar gains a **Pin baseline** button and a **single adaptive "Ask Claude…"
+button** (Edit Selection when text is selected in the preview / Edit Document
+otherwise), plus a minimal right-click → Open Review Preview gateway.
+
+The session opened by classifying `/goal next` → brainstorming, claiming session
+ID **0036** (`--type brainstorming`), and verifying a clean pushed `main`. It then
+oriented from the #43 issue (read via the Gitea API) and the F10 interactive-review
+spec (the architecture F11 extends), and dispatched an Explore agent to produce a
+precise code map of the preview webview, the markdown-it render path, the F4 edit
+seam, `pinDiffBaseline`, and the #41/#42 menu state.
+
+## Pre-state
+
+- `main` clean and pushed; no in-flight sessions (0035 finalized).
+- F10 (`#29`) shipped: clean editor + rendered preview as the single review
+  surface (annotations on/off; ✓/✗ on F4 proposals). `#31` shipped: proposals
+  rendered inline at their resolved anchor.
+- `cowriting.pinDiffBaseline` registered but `when:false` (orphaned since `#34`).
+- Whole-document editing did not exist; "Edit Selection" was an editor
+  context-menu item only.
+
+## Turn-by-turn arc
+
+1. **Classify + claim.** `/goal next` → resume "design #43" → brainstorming.
+   Claimed ID 0036; resolved app facts (content repo present; roadmap = this repo;
+   no BDD corpus); confirmed clean baseline.
+2. **Orient.** Read issue `#43` (Gitea API, P1/feature), the F10 spec
+   (`coauthoring-interactive-review.md`), and prior content-repo specs/plans. Ran
+   an Explore agent → code map: `trackChangesPreview.ts` (webview + annotations
+   toggle round-trip), `trackChangesModel.ts` (markdown-it; `splitBlocksWithRanges`
+   → `BlockWithRange` char offsets; **no** source-line mapping in rendered HTML),
+   `extension.ts editSelection` + `proposalController.propose` + F4 `applyAgentEdit`
+   seam, `diffViewController.pinDiffBaseline` (`when:false`), and the absence of
+   #41/#42 menu entries.
+3. **Surface the forks.** Identified the central design risk — mapping a *rendered
+   preview* selection back to a *source markdown range* — and three load-bearing
+   forks. Asked the operator (AskUserQuestion) with recommendations.
+4. **Forks locked** (all recommended options chosen): (1) **block-level**
+   selection→source mapping via `data-src` block attributes; (2) document edit
+   **diffed into per-hunk F4 proposals**; (3) **#43 lands a minimal right-click
+   gateway**.
+5. **Present design + write spec.** Presented the consolidated F11 design
+   (INV-35/36/37, message bus, slices, testing). Wrote
+   `docs/superpowers/specs/2026-06-12-f11-preview-toolbar-interaction-surface.md`;
+   self-reviewed; verified reused seams exist (`DiffViewController.pin(document)`,
+   `isAuthorable(scheme)`). Committed to `main` (`b2de25f`; rebased over the
+   0036 placeholder push), matching the F9 direct-`design(...)`-commit precedent.
+6. **Finalize.** Submitted the spec to the content repo `specs/` (graduated,
+   `855acec`); updated memory; published this transcript.
+
+## Cut state (final)
+
+- **Spec graduated:** `specs/2026-06-12-f11-preview-toolbar-interaction-surface.md`
+  in `vscode-cowriting-plugin-content` (`855acec`); code-repo working copy at
+  `docs/superpowers/specs/` (`b2de25f` on `main`).
+- `main` clean and pushed. No open PRs (Gitea host — no `gh` PR flow).
+- Memory: added `f11-preview-toolbar-spec-graduated.md` + MEMORY.md pointer.
+
+## Operator plate (decisions & deferrals)
+
+- **Three forks** locked interactively (spec §6.7) — see Summary.
+- **Autonomous low-confidence calls (spec §9, all low-risk):** host `showInputBox`
+  for the instruction (keeps LLM/secrets out of the sealed webview); Pin targets
+  the previewed document via `DiffViewController.pin` (not `activeTextEditor`); new
+  `cowriting.editDocument` command for #42 reuse; `cowriting.pinDiffBaseline`
+  unhidden (`when: editorLangId == markdown`).
+- **Deferred (spec §9):** char-precise sub-block selection mapping; richer #41/#42
+  menu sets; scroll-sync (#32); large-rewrite hunk capping; repo rename (#35).
+
+## Deferred decisions
+
+The autonomous calls above (spec §9) were surfaced at closeout: instruction-prompt
+location, Pin target, the new `editDocument` command, and unhiding the pin command.
+All are low-risk and reversible; none changed the persisted model. No other
+low-confidence calls.
+
+## Next-session prompt
+
+```
+/goal plan-and-execute #43 (F11), per specs/2026-06-12-f11-preview-toolbar-interaction-surface.md
+```
+
+Start with **SLICE-1** (Pin baseline button + unhide `cowriting.pinDiffBaseline`)
+— the immediate win that homes the orphaned command. Read
+`f11-preview-toolbar-spec-graduated.md` and the graduated spec's §6/§7 first.
@@ -0,0 +1,122 @@
+# Session 0037.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-12T13-33 (PST)
+> End: 2026-06-12T14-22 (PST)
+> Type: planning-and-executing
+> Posture: autonomous (yolo)
+> Status: **FINALIZED.**
+>
+> Outcome: **F11 (#43) shipped to main via PR #44.** The rendered review preview's
+> header toolbar is now the primary interaction surface — Pin baseline + a single
+> adaptive Ask-Claude button (Edit Selection ⇆ Edit Document) routing through the
+> existing F4/F6/F3 machinery, block-level selection→source mapping, document
+> edits diffed into per-hunk proposals, an `editor/title` gateway, and the
+> unhidden `pinDiffBaseline`. 208 unit + 9/9 F11 host E2E green. Issue #43 closed;
+> branch deleted; plan archived to the content repo.
+
+## Launch prompt
+
+```
+plan-and-execute #43 (F11), per specs/2026-06-12-f11-preview-toolbar-interaction-surface.md. Start with SLICE-1 (Pin baseline button + unhide pinDiffBaseline) — the immediate win.
+```
+
+## Plan
+
+Plan-and-execute **F11 (#43)** — preview toolbar as the primary interaction
+surface — per `docs/superpowers/specs/2026-06-12-f11-preview-toolbar-interaction-surface.md`.
+Implementation plan archived at `docs/superpowers/plans/2026-06-12-f11-preview-toolbar.md`.
+Posture: autonomous (yolo). Branch `f11-preview-toolbar`.
+
+Five slices (spec §7.2), each green (unit + host E2E) before the next:
+
+- **SLICE-1 — Pin baseline button + reachability** *(this session's lead — the immediate win)*: webview Pin button → `{pinBaseline}` → host `DiffViewController.pin(previewedDoc)`; unhide `cowriting.pinDiffBaseline` (`when: editorLangId == markdown`).
+- **SLICE-2 — Block-offset emission** (INV-36 data layer): pure `data-src-start/end` on every block, both render modes.
+- **SLICE-3 — Edit Document + hunk path** (INV-37 doc half): `diffToHunks`, `runEditAndPropose`, `cowriting.editDocument`.
+- **SLICE-4 — Adaptive Edit Selection** (INV-37 selection half): `selectionchange` label flip + selection→`data-src` resolution → single-range propose.
+- **SLICE-5 — Gateway, edges, tests & docs**: `editor/title` gateway, non-authorable disabling, MANUAL-SMOKE-F11.md, README.
+
+## Session arc
+
+Pre-state: clean `main` at `b2de25f` (after a fast-forward of 2; the F11 spec had
+just graduated in session 0036). #43 = `type/feature` with a graduated Solution
+Design → eligible planning-and-executing anchor (R2/R3).
+
+1. **Claim + setup.** Session gate → `wgl-planning-and-executing`; claimed ID
+   **0037** (no concurrent sessions). Read the F11 spec (found at
+   `docs/superpowers/specs/`, not `specs/`). Wrote the implementation plan
+   (`docs/superpowers/plans/2026-06-12-f11-preview-toolbar.md`), branched
+   `f11-preview-toolbar`. Baseline 197 unit green.
+2. **SLICE-1 (commit 8b9e61a).** TDD: failing E2E (`receiveMessage` seam absent) →
+   extracted `handleWebviewMessage`, added the `pinBaseline` intent →
+   `DiffViewController.pin(previewedDoc)`, the Pin button, theme CSS, unhid
+   `pinDiffBaseline`. Discovered the `undoMarks` E2E flakes (passed on re-run).
+3. **SLICE-2 (1ef9451).** Pure `srcAttr` helper threaded through renderOp /
+   renderReviewOp / renderReview; `renderPlain` switched to per-block bare divs.
+   data-src on every live block in both modes; removed/proposal blocks carry none.
+4. **SLICE-3 (0d1a563).** Pure `diffToHunks` (word-level, coalescing); host
+   `runEditAndPropose` + injectable `editTurn`/`setEditTurnForTest`; the askClaude
+   message + `cowriting.editDocument`; Edit Document button.
+5. **SLICE-4 (03b61ed).** Webview adaptive label (`selectionchange`) +
+   `nearestSrc`/`selectionSrcRange` block-union mapping; host range branch
+   (already shared from SLICE-3). E2E for the selection path.
+6. **SLICE-5 (1564ef5).** `editor/title` gateway; `authorable` render flag +
+   `editControlsEnabled` seam + webview disable; `MANUAL-SMOKE-F11.md`; README.
+7. **Self code review (subagent) → fixes (47cc733).** Caught a **Critical**:
+   pure-insertion hunks were born-orphaned (verified against `anchorer.resolve`).
+   Fixed with `anchorInsertion`; added reconstruct + accept-all E2E coverage;
+   added `turnId`; documented the renderPlain cross-block tradeoff + a
+   characterization test. Ran the isolation experiment proving the `undoMarks`
+   flake is F11-independent.
+8. **Ship.** Pushed; PR **#44** (Gitea API) → merged to `main` (62a2229); #43
+   commented + closed; branch deleted; 208 unit green on main.
+
+Cut state: on `main`, clean, fully pushed. The only red in the host E2E suite is
+the pre-existing `undoMarks` flake (see Deferred decisions); all 9 F11 E2E + 208
+unit are green.
+
+## Next /goal
+
+```
+/goal plan-and-execute #42 then #41 (now unblocked by F11) — expand the right-click menu sets into the preview (Ask Claude to Edit + Open Review Panel), per spec §6.7 fork 3; first capture the pre-existing undoMarks E2E flake as a follow-up bug.
+```
+
+## Deferred decisions
+
+_Autonomous-mode low-confidence calls the driver made and would have
+liked operator input on. Appended as the session runs; surfaced at
+finalize. Empty if none._
+
+- **Pre-existing E2E flake — PROVEN F11-independent, recommend a follow-up issue:**
+  `undoMarks.test.ts` ("F10 #38 — undo of a deletion of baseline text leaves it
+  unattributed") fails on `executeCommand("undo")` not reverting the
+  `applyEdit`-applied changes. Early in the session it failed ~1/3 (passing on
+  re-run); later in the session it failed consistently (machine-load sensitive).
+  **Isolation experiment:** removed all F11 tests, clean rebuild → the undo test
+  still fails identically (44 passing, same `undo restored 'bravo'` at the same
+  line). So it is NOT caused/aggravated by F11 — F11 touches only preview message
+  routing + the pure render layer. Tried a focus+single-undo+poll hardening; it
+  did NOT fix this run (root cause is `undo` not reverting programmatic
+  `WorkspaceEdit`s in test-electron, not mere timing), so I reverted it to keep
+  the F11 PR scoped. **Recommend a dedicated follow-up issue:** redesign the undo
+  test to not depend on `executeCommand("undo")` over `applyEdit` (e.g. drive the
+  attribution branch directly, or type real edits). All 9 F11 E2E tests are green.
+- **SLICE-2 off-mode wrapping (autonomous call):** `renderPlain` now wraps each
+  block in a bare `<div data-src-start/end>` (no `cw-` class) rather than
+  rendering the whole document in one markdown pass. This makes the off/clean
+  preview a selection→source surface (INV-36) and matches `renderReview`'s
+  long-standing per-block rendering. Trade-off (raised in code review): cross-block
+  markdown constructs separated by blank lines (reference-link defs, footnotes)
+  don't resolve in off-mode — a regression for off-mode specifically, but it makes
+  both modes consistent (on-mode already had it). Inherent to the operator-locked
+  block-level mapping (§6.7). Kept per INV-36 ("both modes"); documented with a
+  docstring note + a characterization test. A cross-block-fidelity improvement
+  (markdown-it source maps) is a possible follow-up, not a change to the locked
+  decision.
+- **Code review run (self, via subagent) before merge:** found 1 Critical (pure-
+  insertion hunks born-orphaned — could never be accepted), 2 Important (the
+  cross-block renderPlain regression above; sequential multi-hunk accept untested),
+  3 Minor (no turnId; test-seam recompute; webview disable not auto-tested). Fixed
+  the Critical (anchorInsertion + reconstruct/accept-all tests), the turnId, and
+  added accept-all + characterization coverage; accepted/documented the cross-block
+  tradeoff. Commit `47cc733`.
@@ -0,0 +1,88 @@
+# Session 0038.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-12T15-55 (PST)
+> End: 2026-06-12T16-06 (PST)
+> Type: planning-and-executing
+> Posture: autonomous (yolo)
+> Status: **FINALIZED**
+
+## Pre-state
+
+`main` clean and pushed (HEAD `21df670`, behind origin by 1 — fast-forwarded to
+`de53305` the session-claim commit). No in-flight sessions. Open backlog: #42
+(feature, Ask Claude to Edit Document), #41 (story, right-click → Open Review
+Panel); both filed in capture session 0035 as F11 follow-ups.
+
+## Launch prompt
+
+`/goal next` → stored goal "plan-and-execute #42 (Ask Claude to Edit Document),
+then #41" (from session 0037 finalize).
+
+## Plan
+
+**Gate decision (§4.3 R3):** #42 is `type/feature` with no Solution Design — an
+ineligible anchor for plan-and-execute. Operator chose: **plan-and-execute #41
+now (eligible story); brainstorm #42's design next.** So this session does #41
+only; #42's design becomes the next goal.
+
+**#41 — Open Cowriting Review Panel from the markdown file/tab right-click menu**
+(story, P2):
+- `explorer/context` menu item for `.md` → `cowriting.showTrackChangesPreview`.
+- `editor/title/context` menu item for markdown tabs → same command.
+- Command accepts the right-clicked resource URI (open clicked doc, not just
+  active editor).
+- Menu label reads "Open Cowriting Review Panel" (least-churn option).
+- E2E coverage: menu contributed + opens preview.
+
+### Outcome
+
+**#41 SHIPPED** to main (PR #45, merged + branch deleted; issue auto-closed).
+- `explorer/context` + `editor/title/context` menus (markdown-gated) →
+  `cowriting.showTrackChangesPreview`; command retitled "Open Cowriting Review
+  Panel"; command resolves the *clicked* doc (opens it if not already open),
+  no-arg path unchanged.
+- 208 unit + 7 new E2E (`reviewPanelMenu.test.ts`); full host E2E suite green
+  (66 passing across both workspaces); typecheck clean.
+
+**#42 NOT done this session** — `type/feature` with no Solution Design is an
+ineligible plan-and-execute anchor (§4.3 R3). Operator chose: brainstorm #42's
+design next. → **Next /goal:** brainstorm #42 (Ask Claude to Edit Document)
+Solution Design, then plan-and-execute #42.
+
+## Deferred decisions
+
+- #42 gate (R3): asked the operator rather than auto-overriding; they chose
+  "#41 now, brainstorm #42 next." Not low-confidence — recorded for the
+  finalize report.
+- Pre-existing F11-independent `undoMarks` E2E flake noted in session 0037 still
+  uncaptured as a follow-up issue (out of scope for #41).
+
+## Session arc
+
+1. Claimed session 0038 (peek → no in-flight → atomic push). Fast-forwarded
+   `main`. Fetched #42 + #41 from the Gitea API.
+2. Ran the §4.3 issue-type gate: #42 (feature, no design) ineligible for
+   plan-and-execute; #41 (story) eligible. Surfaced the fork; operator chose
+   "#41 now, brainstorm #42 next."
+3. Implemented #41: read `trackChangesPreview.ts` (the `showTrackChangesPreview`
+   command already took an optional `uri` but only searched *open* docs) +
+   `package.json` menus. Fixed the command to open the clicked doc; added
+   `explorer/context` + `editor/title/context`; retitled the command; updated
+   README.
+4. Wrote `reviewPanelMenu.test.ts` (7 E2E). Typecheck + 208 unit + full host
+   E2E green.
+5. Branch → commit → push → PR #45 → merged (branch deleted); #41 auto-closed.
+6. Finalized: memory updated, transcript published.
+
+## Next-session prompt
+
+```
+/goal brainstorm #42 (Ask Claude to Edit Document) Solution Design, then plan-and-execute #42
+```
+
+**Next /goal:** brainstorm #42 (Ask Claude to Edit Document) Solution Design,
+then plan-and-execute #42. Design forks to settle: whole-doc instruction
+gathering; one large proposal vs several; large-document token-limit behavior.
+`cowriting.editDocument` + `runEditAndPropose(doc, {kind:"document"})` already
+exist (F11 prep) — #42 is mostly menu/tab wiring + those decisions on top.
@@ -0,0 +1,74 @@
+# Session 0039.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-12T16-07 (PST)
+> End: 2026-06-12T17-02 (PST)
+> Type: capture (tracked-lite)
+> Posture: careful (capture default)
+> Status: **FINALIZED**
+
+## Launch prompt
+
+`/wgl-capture We need a way to accept all of Claude's edits`
+
+## Pre-state
+
+Opened immediately after finalizing session 0038. `main` clean + pushed.
+Backlog open: #42 (feature, Ask Claude to Edit Document). No existing
+accept-all draft in the content repo.
+
+## What was captured
+
+**One ask → one issue filed.**
+
+**#46** — `type/feature`, `priority/P2` —
+"Accept all of Claude's pending proposals on the current document in one gesture."
+https://git.benstull.org/benstull/vscode-cowriting-plugin/issues/46
+
+Confirmed gap from the code: accept is **per-proposal only** today
+(`cowriting.acceptProposal` → `ProposalController.acceptById`; F10/F11 preview ✓
+per `cw-proposal` block). No bulk accept. Felt most after F11 "Edit Document"
+(#42), where one whole-document rewrite fans out into N per-hunk proposals.
+
+## Decisions (capture forks)
+
+- **Scope of "all" = all pending proposals on the current document** (operator
+  choice) — not per-turn, not cross-document. Safe apply-order; orphaned /
+  unresolved proposals **skipped + reported**, never force-applied.
+- **Type = `type/feature`** (operator choice). The load-bearing fork is
+  apply-order + re-anchoring: each accept mutates the buffer and re-resolves the
+  remaining proposals' anchors, so bulk-accept needs a correct order /
+  re-resolve-between strategy. Design step also settles confirmation y/n, button
+  placement (preview header vs F11 toolbar), and applied-vs-skipped surfacing.
+- Non-goals recorded: no reject-all, no cross-document/workspace accept-all, no
+  per-turn scope, no new accept seam (reuse F4 `acceptById` in a loop; webview
+  posts an `acceptAll` intent — sealed-webview INV-21/35).
+
+## Session arc
+
+1. Claimed tracked-lite session 0039 (`--type capture`); fast-forwarded `main`.
+2. Resolved the content repo (`vscode-cowriting-plugin-content`); cloned + listed
+   existing drafts (no accept-all duplicate).
+3. Inspected the code's accept surface → confirmed per-proposal-only, no bulk.
+4. Reflected the ask; asked two scope/sizing questions → current-doc + feature.
+5. Drafted `issues/accept-all-proposals.md` (full §5 feature template).
+6. Operator approved → ensured labels → filed **#46**.
+7. Finalized: memory updated, transcript published. Draft left uncommitted in the
+   content-repo working tree (INV-8 — Author to publish/discard).
+
+## Deferred decisions
+
+- Capture scripts 401 with the default token service; needed
+  `WGL_CAPTURE_TOKEN_SERVICE=wgl-gitea-issues-readwrite-token-git.benstull.org`
+  (host-specific). Worth a plugin-feedback follow-up if it recurs.
+
+## Next-session prompt
+
+Unchanged from 0038 — the next coding goal stands:
+
+```
+/goal brainstorm #42 (Ask Claude to Edit Document) Solution Design, then plan-and-execute #42
+```
+
+#46 (accept-all) is now in the backlog alongside #42; either could be the next
+brainstorming target.
@@ -0,0 +1,81 @@
+# Session 0040.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-12T17-03 (PST)
+> End: 2026-06-12T17-21 (PST)
+> Type: capture (tracked-lite)
+> Posture: careful (capture default)
+> Status: **FINALIZED**
+
+## Launch prompt
+
+`/wgl-capture When Claude suggests changes, a single change should be for a
+paragraph (or header, or bulletpoint), not individual words. It's too much to
+review`
+
+## Pre-state
+
+Third session of the run (after 0038 ship + 0039 capture). `main` clean +
+pushed. Backlog: #42 (Edit Document), #46 (accept-all, filed 0039).
+
+## What was captured
+
+**One ask → one issue filed.**
+
+**#47** — `type/feature`, `priority/P1` —
+"Block-granularity proposals: one change per paragraph/header/bullet, not per word."
+https://git.benstull.org/benstull/vscode-cowriting-plugin/issues/47
+
+Confirmed from code: F11 **Ask Claude to Edit Document** mints proposals at **word
+granularity** — `diffToHunks` (src/trackChangesModel.ts) uses `diffWordsWithSpace`
+and flushes a hunk on every unchanged token, so each changed word is its own ✓/✗
+proposal (INV-37 "per-hunk"). A light copy-edit pass explodes into dozens of tiny
+proposals → "too much to review."
+
+## Decisions (capture forks)
+
+- **Scope = document edits only** (operator choice) — the `diffToHunks` fan-out.
+  Selection edits are already a single proposal, so untouched.
+- **Type = `type/feature`** (operator choice). Supersedes F11 **INV-37 per-hunk →
+  per-block**. Real forks recorded for the design step: block taxonomy (list items
+  vs whole lists, tables, blockquotes, nesting); anchoring inserted blocks (a
+  block-level analogue of today's word-oriented `anchorInsertion`); and the
+  **attribution tradeoff** — accepting a block attributes the *whole* block to
+  Claude, including unchanged words.
+- Reuse the existing block splitter (`diffBlocks`; code/mermaid fences atomic,
+  INV-23). Intra-block `<ins>`/`<del>` **rendering is unchanged** — only the
+  decision *unit* becomes the block. `priority/P1` (directly removes the stated
+  pain).
+
+## Session arc
+
+1. Claimed tracked-lite session 0040 (`--type capture`); fast-forwarded `main`.
+2. Inspected `diffToHunks` → confirmed word-level granularity is the cause.
+3. Reflected the ask; asked scope + sizing → document-edits + feature.
+4. Drafted `issues/block-granularity-proposals.md` (full §5 feature template).
+5. Operator approved → filed **#47** (labels already ensured this run).
+6. Finalized: memory updated, transcript published. Draft left uncommitted in the
+   content-repo working tree (INV-8).
+
+## Deferred decisions
+
+_None._
+
+## Backlog observation
+
+#42 (Ask Claude to Edit Document), #46 (accept-all), and #47 (block granularity)
+form one **edit-flow cluster** around the Ask-Claude / document-edit loop. They
+interact (block granularity changes how many proposals exist; accept-all changes
+how they're taken; #42 is the entry point that produces them) and are best
+**brainstormed / sequenced together** rather than designed in isolation.
+
+## Next-session prompt
+
+The standing coding goal is unchanged; consider folding the cluster in:
+
+```
+/goal brainstorm the Ask-Claude document-edit cluster (#42 Edit Document, #47 block-granularity proposals, #46 accept-all) and sequence/design it, starting with #42
+```
+
+(Or keep the narrower `/goal brainstorm #42 … then plan-and-execute #42` and take
+#47/#46 after.)
@@ -0,0 +1,83 @@
+# Session 0041.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-12T18-17 (PST) · End: 2026-06-13T07-16 (PST)
+> Type: brainstorming
+> Posture: autonomous (yolo)
+> Goal: brainstorm the Ask-Claude document-edit cluster (#42, #47, #46) and
+> sequence/design it, starting with #42
+> Outcome: **GRADUATED** — one combined Solution Design
+> `specs/coauthoring-document-edit-flow.md` (content repo, `status: graduated`,
+> commit `9d016e8`), sequencing #42 → #47 → #46.
+
+## Launch prompt
+
+`/goal brainstorm the Ask-Claude document-edit cluster (#42, #47, #46) and
+sequence/design it, starting with #42`
+
+## Plan
+
+Brainstorming session — produce a Solution Design for the **Ask-Claude
+document-edit flow**, the cluster of #42 (Ask Claude to Edit Document — entry
+points), #47 (block-granularity proposals), #46 (accept-all). Deliverable: one
+combined Solution Design (content repo `specs/`), sequencing the three as a
+delivery plan **#42 (reach) → #47 (review) → #46 (accept)**.
+
+## Arc
+
+1. **Init.** Claimed session 0041 (`--type brainstorming`), clean pushed `main`
+   baseline, seeded the Plan block. Resolved app repos (roadmap = framework repo;
+   content = `vscode-cowriting-plugin-content`; no BDD corpus).
+2. **Read the cluster.** Read the three capture drafts (#42/#47/#46) in the
+   content repo `issues/`. Then ground in code: `package.json` menus/commands,
+   `trackChangesModel.ts` (`splitBlocks`/`diffBlocks`/`diffToHunks`/
+   `anchorInsertion`/`renderReview`), `trackChangesPreview.ts`
+   (`editDocument`/`askClaude`/`runEditAndPropose`), `proposalController.ts`
+   (`acceptById`/`accept`→`applyAgentEdit`/`listProposals`).
+3. **Key reframe.** F11 (session 0037) **already shipped** the document-edit path
+   — `cowriting.editDocument` → `runEditAndPropose({kind:"document"})` →
+   `diffToHunks` (word-level, INV-37). So #42 is mostly DONE (remaining = menu
+   wiring); #47 swaps the document-branch diff to block-level; #46 loops the
+   existing accept seam.
+4. **Found a spec-hygiene gap.** The F11 (#43) Solution Design was never graduated
+   to `specs/` — INV-35/36/37 live only in code comments + the F11 issue draft.
+   Logged as OQ-2 (follow-up), restated the depended-on F11 invariants in the new
+   spec.
+5. **Locked two forks with the operator** (AskUserQuestion):
+   - **D-1** — per-block proposals **preserve unchanged-span attribution** (block =
+     decision unit, word = attribution unit; intra-block `diffToHunks` sub-diff at
+     accept; INV-40). *Not* whole-block→Claude.
+   - **D-2** — ship as **one combined Solution Design**, not three.
+6. **Wrote the spec.** `specs/coauthoring-document-edit-flow.md` — business →
+   product → engineering, new invariants INV-38..42 (INV-39 supersedes INV-37),
+   delivery plan SLICE-1/2/3 = #42/#47/#46, traceability, decisions log. Self-
+   reviewed (no placeholders, consistent, scoped). Did **not** invoke
+   `writing-plans` (wgl-brainstorming override — the artifact is the spec,
+   submitted at finalize; implementation is a downstream session).
+7. **Finalize.** Submitted the spec `--status graduated`; memory updated.
+
+## Deferred decisions
+
+- **D-3 (low-confidence driver call)** — sequenced **#42 → #47 → #46** per the
+  goal's "starting with #42," **despite #47 carrying the P1 value** ("too much to
+  review" is the sharpest pain). The three are code-independent so the order is
+  reversible (§7.1); #47 and #42 could swap with no rework if the operator wants
+  the P1 review fix to land first. Surfaced at closeout; logged as D-3 in the
+  spec §9.
+- D-4 (driver) block taxonomy = exactly `splitBlocks`'s units; D-5 (driver)
+  accept-all needs no confirmation, button shown ≥2 pending. Both in spec §9.
+
+## Loose ends → next session
+
+- **OQ-2** — graduate the F11 (#43) spec to `specs/` for a clean invariant ledger
+  (INV-35/36/37 currently code-only).
+- **OQ-1** — whether selection edits should also preserve-unchanged-span
+  attribution (out of #47 scope).
+- Content-repo local working tree carries pre-existing untracked drafts/specs
+  (prior sessions / the Author's) — not landed by this session.
+
+## Next /goal
+
+```
+/goal plan-and-execute #42 (SLICE-1 of the document-edit flow — Ask-Claude entry-point menu wiring), per specs/coauthoring-document-edit-flow.md
+```
@@ -0,0 +1,62 @@
+# Session 0042.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-12T21-04 (PST) · End: 2026-06-13T07-16 (PST)
+> Type: capture (tracked-lite)
+> Posture: autonomous (yolo)
+> Goal: capture "When I pin the baseline, there should be no annotations in the
+> review panel"
+> Outcome: filed **#48** (`type/story`, `priority/P2`).
+
+## Launch prompt
+
+`/wgl-capture When I pin the baseline, there should be no annotations in the
+review panel`
+
+Opened mid-session, concurrent with the in-flight brainstorming session 0041
+(only that — my own — session in flight; warned + proceeded per autonomous).
+
+## Arc
+
+1. **Claim.** Claimed session 0042 (`--type capture`, tracked-lite). App +
+   content repo already resolved from 0041.
+2. **Ground the behavior.** Read `renderReview`/`colorByAuthor`/`renderReviewOp`
+   in `trackChangesModel.ts`: an **unchanged** block renders as
+   `cw-blk cw-unchanged` with `colorByAuthor` applied (line ~526). So after a Pin
+   baseline (baseline == current → every block unchanged), the change axis is
+   empty but **F3 authorship coloring still paints every block** — and the panel
+   *does* refresh on pin (`onDidChangeBaseline`), so it's not a stale render. The
+   author colors are the "annotations."
+3. **Scope fork (AskUserQuestion).** Operator chose **pin → fully clean panel**
+   (zero-diff state: change marks AND authorship colors gone), **not** the broader
+   "authorship never colors unchanged text." → a single, design-free increment.
+4. **Type.** Would be `type/bug`, but `capture-issues.sh` accepts only
+   epic|feature|story|task (the gap plugin feedback #124 tracks — not re-filed).
+   Filed as the accurate fileable type `type/story`, P2.
+5. **Draft + approval + file.** Authored
+   `issues/pin-baseline-clears-review-annotations.md`; operator **approved**;
+   ensured labels (`--host git.benstull.org`,
+   `WGL_CAPTURE_TOKEN_SERVICE=wgl-gitea-issues-readwrite-token-git.benstull.org`);
+   filed **#48** —
+   https://git.benstull.org/benstull/vscode-cowriting-plugin/issues/48
+
+## Deferred decisions
+
+- **Type compromise (driver, tooling-forced)** — filed as `type/story` though the
+  observation is bug-flavored (surprising behavior on a primary gesture);
+  `capture-issues.sh` can't type `bug`. The bug-typing gap is already tracked as
+  plugin feedback **#124**; not re-filed.
+
+## Loose ends
+
+- Draft `issues/pin-baseline-clears-review-annotations.md` left **uncommitted** in
+  the content repo working tree (the Author's to publish or discard — capture
+  INV-8).
+
+## Next /goal
+
+(Continuation is the 0041 handoff.)
+
+```
+/goal plan-and-execute #42 (SLICE-1 of the document-edit flow — Ask-Claude entry-point menu wiring), per specs/coauthoring-document-edit-flow.md
+```
@@ -0,0 +1,92 @@
+# Session 0043.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-13T07-18 (PST)
+> Type: planning-and-executing
+> End: 2026-06-13T07-38 (PST)
+> Status: **FINALIZED.**
+> Posture: autonomous (yolo)
+
+## Launch prompt
+
+`/goal plan-and-execute #42 (SLICE-1 of the document-edit flow — Ask-Claude
+entry-point menu wiring), per specs/coauthoring-document-edit-flow.md`
+
+## Plan
+
+**SLICE-1 — #42 (reach)** of the document-edit flow Solution Design
+(`specs/coauthoring-document-edit-flow.md`, §7.2). Anchor: Feature #42
+(`type/feature`); design graduated (session 0041) → §4.3 R3 satisfied.
+
+Goal (INV-38): "Ask Claude to Edit" reachable from the editor **body** and the
+editor **tab**, selection-aware (selection → `editSelection`; no selection →
+`editDocument`), both gated to markdown/authorable docs, both routing through the
+single `runEditAndPropose` path. Smallest increment — command already exists.
+
+Tasks:
+1. `package.json` — add `editSelection` + `editDocument` to `editor/context`
+   (selection-aware `when`, markdown+authorable) and `editor/title/context`
+   (selection-aware, `resourceLangId == markdown`); keep titles.
+2. `trackChangesPreview.ts` — `cowriting.editDocument` accepts an optional tab
+   `uri` arg (mirror `showTrackChangesPreview`'s #41 resolution): open/resolve the
+   clicked doc, else fall back to the active editor.
+3. E2E (`test/e2e/suite/`) — menu entries present + selection-aware + markdown-gated;
+   `editDocument(uri)` resolves the tab doc and produces a document-scoped proposal.
+
+No model change, no new persisted artifact. No deploy pipeline (VS Code extension).
+
+## Results
+
+**SLICE-1 (#42, reach) shipped to `main`** — PR
+[#49](https://git.benstull.org/benstull/vscode-cowriting-plugin/pulls/49)
+(merged), issue #42 closed.
+
+- `package.json` — `editSelection` + `editDocument` added to `editor/context`
+  (selection-aware, `editorLangId == markdown` + `file`/`untitled`) and
+  `editor/title/context` (selection-aware, `resourceLangId == markdown`);
+  existing `editor/context` `editSelection` entry markdown-gated to match INV-38.
+- `src/trackChangesPreview.ts` — `cowriting.editDocument` now accepts the clicked
+  tab's resource `Uri` (opens it if needed), falling back to the active editor;
+  mirrors `showTrackChangesPreview`'s #41 resolution.
+- `test/e2e/suite/f12Reach.test.ts` — 4 new host E2E (menu wiring declarative +
+  tab-URI targeting + no-arg fallback).
+- `docs/MANUAL-SMOKE-F12.md` — SLICE-1 reach smoke steps.
+
+Verification: `tsc --noEmit` clean; **208 unit** green; **65/5 host E2E** green
+(main suite up from 61 → 65 with the 4 new tests).
+
+**Next:** SLICE-2 — #47 (review, **P1**): document edits propose per changed
+block (`diffToBlockHunks`, INV-39 supersedes INV-37; word-precise intra-block
+attribution INV-40; block-insertion anchoring INV-41). Then SLICE-3 — #46
+(accept-all, INV-42).
+
+## Session arc
+
+1. **Gate + claim.** Classified the `/goal` as planning-and-executing; claimed
+   session 0043 (peek showed nothing in flight). Baseline: local `main` was 4
+   behind `origin/main` → fast-forwarded clean.
+2. **Anchor gate (§4.3 R3).** #42 is `type/feature`; its design is the graduated
+   combined Solution Design `specs/coauthoring-document-edit-flow.md` (session
+   0041) → R3 satisfied, proceed.
+3. **Read the spec**, scoped SLICE-1 (reach). Explored the code: `package.json`
+   menus, `editDocument`/`editSelection` handlers, the F11 E2E harness (`setEditTurnForTest`).
+4. **TDD.** Wrote `f12Reach.test.ts` first (red), then wired the menus + tab-URI
+   resolution (green).
+5. **Verified** (tsc clean, 208 unit, 65/5 E2E), self-reviewed the diff, shipped
+   via PR #49 (merged), closed #42, updated memory, checkpoint-published the
+   transcript.
+
+## Deferred decisions
+
+_Autonomous-mode low-confidence calls the driver made and would have
+liked operator input on. Appended as the session runs; surfaced at
+finalize. Empty if none._
+
+- **Markdown-gating `editSelection`'s `editor/context` entry** (driver call,
+  low-confidence). Spec §5/INV-38 say both Ask-Claude entries are markdown-gated,
+  but F8 made the `editSelection` *command* work on any authorable doc and its
+  current right-click entry has no `editorLangId == markdown` gate. Followed the
+  spec: gated the **menu** entries to markdown while leaving the command handlers'
+  behavior intact (palette still works on any authorable file). Removes the
+  body right-click Ask-Claude-Edit-Selection on non-markdown files — acceptable
+  since the proposal review surface (F10 preview) is markdown-only.
@@ -0,0 +1,127 @@
+# Session 0044.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-13T07-40 (PST)
+> Type: planning-and-executing
+> End: 2026-06-13T08-04 (PST)
+> Status: **FINALIZED.**
+> Posture: autonomous (yolo)
+
+## Launch prompt
+
+```
+/goal plan-and-execute #47 (SLICE-2 of the document-edit flow — document edits propose per changed block: diffToBlockHunks INV-39 supersedes INV-37, word-precise intra-block attribution INV-40, block-insertion anchoring INV-41), per specs/coauthoring-document-edit-flow.md
+
+```
+
+## Plan
+
+**SLICE-2 — #47 (review, P1)** of the document-edit flow
+(`specs/coauthoring-document-edit-flow.md` §7.2). Anchor: Feature #47
+(`type/feature`); covered by the graduated combined Solution Design → §4.3 R3.
+
+Document rewrites propose **one F4 proposal per changed block** (the unit a human
+reviews), but each accept reconciles attribution at **word** granularity (the unit
+F3 records). INV-39 supersedes INV-37 for document edits; INV-40 word-precise
+intra-block attribution; INV-41 block-insertion anchoring. Selection edits
+unchanged.
+
+Tasks:
+1. `model.ts` — add optional `granularity?: "block" | "single"` to `Proposal`
+   (additive, back-compat: absent ⇒ `"single"`). INV-39/§6.3.
+2. `proposalModel.ts` — `addProposal` opts gains `granularity`, threaded onto the
+   proposal (omitted when absent).
+3. `trackChangesModel.ts` — new pure `diffToBlockHunks(currentText,
+   rewrittenText): EditHunk[]`: split both sides into the existing block units
+   (`splitBlocksWithRanges` for source ranges + `splitBlocks` keys), diff blocks
+   (reuse `diffArrays` keying like `diffBlocks`), emit ONE `EditHunk` per
+   changed/added/removed block — changed → `[block.start,end)`→rewrite; fences
+   atomic (whole-fence hunk); inserted → anchored to adjacent boundary (INV-41,
+   reuse the `anchorInsertion` idea); unchanged → none. Same `EditHunk` shape as
+   `diffToHunks` (which is RETAINED as the intra-block sub-diff engine).
+4. `trackChangesPreview.ts` — `runEditAndPropose` **document branch** uses
+   `diffToBlockHunks` and tags each minted proposal `granularity:"block"`.
+5. `proposalController.ts` — `accept`: when `granularity === "block"`, re-resolve
+   the block text, `diffToHunks(blockText, replacement)` → per-run word sub-diff,
+   apply **one `applyAgentEdit` per changed run, descending offset** so only the
+   runs Claude changed land Claude-attributed; unchanged spans keep prior author
+   (INV-40). Non-block keeps the single `applyAgentEdit`.
+6. Tests: unit (`diffToBlockHunks` + INV-40 controller) + host E2E (#47 cases in
+   §6.8); update the existing f11 document-path E2E (now per-block, supersedes
+   INV-37 word-level expectation); `docs/MANUAL-SMOKE-F12.md` §2.
+
+**Design note (seam constraint):** `pendingEdits.matchEvent` resolves ONE
+registration per change event, so INV-40's per-run attribution is implemented as
+**sequential** `applyAgentEdit` calls (descending offset), not a single
+multi-replace `WorkspaceEdit`. The spec's "one undo-grouped edit" wording is thus
+approximated as N undo steps per block accept — see Deferred decisions.
+
+No new persisted artifact; no deploy pipeline (VS Code extension).
+
+## Results
+
+**SLICE-2 (#47, review, P1) shipped to `main`** — PR
+[#50](https://git.benstull.org/benstull/vscode-cowriting-plugin/pulls/50)
+(merged), issue #47 closed.
+
+- `trackChangesModel.ts` — new pure `diffToBlockHunks` (block-key alignment via
+  `diffArrays`/`diffBlocks` keying; isolated changed block → block-aligned hunk →
+  rewritten raw; fences atomic INV-23; insert/delete → exact gap-span hunk;
+  zero-width gap-span anchored INV-41). Split `diffToHunks` into raw
+  `wordEditHunks` + anchoring wrapper (fixes a latent overlap bug under batch apply).
+- `trackChangesPreview.ts` — document branch uses `diffToBlockHunks`, tags
+  `granularity:"block"`.
+- `model.ts`/`proposalModel.ts` — additive optional `Proposal.granularity`.
+- `proposalController.ts` — `acceptBlock`: intra-block word sub-diff
+  (`wordEditHunks`) → one `applyAgentEdit` per changed run, descending offset
+  (INV-40 word-precise attribution; unchanged spans keep prior author).
+- Tests: `diffToBlockHunks` unit (reconstruction/fence/add-remove); `f12Review`
+  host E2E (M→M, unchanged→none, fence atomic, INV-40, INV-41); updated f11
+  document-path E2E to per-block; `MANUAL-SMOKE-F12.md` §2.
+
+Verification: `tsc --noEmit` clean; **214 unit** green; **69/5 host E2E** green
+(main suite 65 → 69 net with the new F12 review tests).
+
+**Next:** SLICE-3 — #46 (accept-all, INV-42): `acceptAllProposals` (controller,
+descending-order, orphan-skip + report) + `cowriting.acceptAllProposals` command
+ preview toolbar **Accept all** button (`{type:"acceptAll"}` intent, shown ≥2
+pending).
+
+## Session arc
+
+1. **Claim.** Resumed via `/goal next` (after session 0043 shipped #42); claimed
+   session 0044 for #47. Clean `main` baseline.
+2. **Anchor gate.** #47 (`type/feature`) covered by the graduated combined design
+   → §4.3 R3 satisfied.
+3. **Deep code read.** `splitBlocks*`/`diffBlocks`/`diffToHunks`/`EditHunk`,
+   `runEditAndPropose`, `proposalController.accept`/`applyAgentEdit`, and crucially
+   `pendingEdits.matchEvent` (one registration per change event → drove the INV-40
+   sequential-apply decision).
+4. **TDD `diffToBlockHunks`.** First attempt (coarsen word-hunks) failed a
+   reconstruction case — discovered `diffToHunks` can emit OVERLAPPING hunks
+   (anchorInsertion over-absorbs). Switched to block-key alignment with
+   replacement-from-rewritten-raw → green.
+5. **Model + wiring + INV-40 accept.** Added `granularity`; document branch →
+   `diffToBlockHunks`; `acceptBlock` intra-block sub-diff. The overlap bug then
+   resurfaced in `acceptBlock` (batch apply) → split `diffToHunks` into raw
+   `wordEditHunks` (disjoint) + anchoring wrapper; `acceptBlock` uses raw → green.
+6. **Verified + shipped** (214 unit, 69/5 E2E), self-reviewed, PR #50 merged,
+   #47 closed, memory + transcript updated.
+
+## Deferred decisions
+
+_Autonomous-mode low-confidence calls the driver made and would have
+liked operator input on. Appended as the session runs; surfaced at
+finalize. Empty if none._
+
+- **INV-40 undo granularity** (driver call). Spec §6.2/§6.7 say accepting a block
+  proposal should be "one undo-grouped WorkspaceEdit." But the F3 seam's
+  `pendingEdits.matchEvent` matches exactly ONE registration per change event, so a
+  single `WorkspaceEdit` with N independent run-replaces can't carry per-run
+  attribution (the multi-hunk event fails to match any single registration → falls
+  through to human). Implemented INV-40 as **sequential per-run `applyAgentEdit`
+  calls, descending offset** — correct word-precise attribution, reuses the seam
+  with zero attribution-core risk, but a block accept is N undo steps rather than
+  one. Single-undo-step would require extending `matchEvent`/`onDidChange` to
+  consume multiple registrations per event (riskier F3 change) — left as a possible
+  follow-up.
@@ -0,0 +1,98 @@
+# Session 0045.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-13T08-04 (PST)
+> Type: planning-and-executing
+> End: 2026-06-13T08-17 (PST)
+> Status: **FINALIZED.**
+> Posture: autonomous (yolo)
+
+## Launch prompt
+
+```
+/goal plan-and-execute #46 (SLICE-3 of the document-edit flow — accept-all, INV-42), per specs/coauthoring-document-edit-flow.md
+
+```
+
+## Plan
+
+**SLICE-3 — #46 (accept)** of the document-edit flow
+(`specs/coauthoring-document-edit-flow.md` §7.2, INV-42). Final slice — completes
+reach→review→**accept**. Anchor: Feature #46 (`type/feature`); covered by the
+graduated combined design → §4.3 R3.
+
+A single **Accept all** gesture applies every pending proposal on the current
+document through the existing F4 `acceptById` seam (block proposals take the INV-40
+path automatically), in a re-anchor-safe (descending) order, **skipping** orphans
+and **reporting** applied-vs-skipped. Batched application of the existing path — no
+new accept mechanism; the webview posts intent only (INV-35).
+
+Tasks:
+1. `proposalController.ts` — `acceptAllProposals(document): Promise<{applied,
+   skipped}>`: snapshot pending, sort descending by resolved anchor start,
+   `acceptById` each **silently** (no per-item orphan warning), tally
+   applied/skipped (orphans counted skipped). Add a `silent` opt to the accept
+   path so the batch suppresses per-proposal warnings.
+2. `extension.ts` — `cowriting.acceptAllProposals` command (active doc) → reports
+   applied-vs-skipped via a status message.
+3. `trackChangesPreview.ts` — `ToolbarMsg += {type:"acceptAll"}`;
+   `handleWebviewMessage` routes it → `proposals.acceptAllProposals(document)` +
+   report.
+4. `media/preview.ts` + `shellHtml` — an **Accept all** toolbar button shown when
+   `summary.proposals >= 2` (and authorable), posting `{type:"acceptAll"}`.
+5. `package.json` — register the `cowriting.acceptAllProposals` command (palette,
+   markdown-gated).
+6. Tests: host E2E (N pending → all applied + cleared; orphan skipped + reported;
+   button hidden < 2 pending); `MANUAL-SMOKE-F12.md` §3.
+
+No new persisted artifact; no deploy pipeline (VS Code extension).
+
+## Results
+
+**SLICE-3 (#46, accept) shipped to `main`** — PR
+[#51](https://git.benstull.org/benstull/vscode-cowriting-plugin/pulls/51)
+(merged), issue #46 closed. **Completes the document-edit-flow cluster
+(#42 reach + #47 review + #46 accept).**
+
+- `proposalController.ts` — `acceptAllProposals(document)` → `{applied, skipped}`
+  (descending order, orphan-skip); `accept`/`acceptById` `silent` opt for the batch.
+- `trackChangesPreview.ts` — `ToolbarMsg += {type:"acceptAll"}` → public
+  `acceptAll(document)` (batch + report).
+- `extension.ts`/`package.json` — `cowriting.acceptAllProposals` command
+  (active doc, markdown-gated palette).
+- `media/preview.ts` + shell — "✓✓ Accept all" toolbar button (intent; shown
+  ≥2 pending, authorable, on-state).
+- `trackChangesModel.ts` — `diffToBlockHunks` fix: emit one block-aligned hunk
+  per changed block **even when adjacent** (changed blocks are 1:1 anchors;
+  gap-spans only cover add/remove runs) — caught by the accept-all E2E (3 adjacent
+  changed blocks were collapsing to 1 proposal).
+- `f12Accept.test.ts` E2E + `MANUAL-SMOKE-F12.md` §3.
+
+Verification: `tsc --noEmit` clean; **214 unit** green; **73/5 host E2E** green
+(main suite 69 → 73 with the new accept-all tests).
+
+**Next:** the cluster is complete; no in-flight next step. Open backlog includes
+#48 (pin → fully clean review panel, story P2) and the OQ-2 F11 (#43) spec
+graduation (hygiene). A natural hand-back point for operator direction.
+
+## Session arc
+
+1. **Claim.** Continued the rolling `next` goal after 0044 shipped #47; claimed
+   session 0045 for #46. Clean `main`.
+2. **Plan + implement.** Controller `acceptAllProposals` (+ `silent` accept opt);
+   `acceptAll` intent route + public method; command + package.json; "Accept all"
+   toolbar button (≥2-pending gating in the sealed webview).
+3. **E2E caught a real bug.** The accept-all "3 adjacent changed blocks" case
+   returned 1 proposal, not 3 — `diffToBlockHunks` was merging adjacent changed
+   blocks into one gap-span run. Fixed by treating `changed` blocks as 1:1 anchors
+   (each → its own block-aligned hunk; gap-spans only span add/remove runs).
+4. **Verified + shipped** (214 unit, 73/5 E2E), self-reviewed, PR #51 merged,
+   #46 closed → **cluster complete**; memory + transcript updated.
+
+## Deferred decisions
+
+_Autonomous-mode low-confidence calls the driver made and would have
+liked operator input on. Empty if none._
+
+- _No low-confidence calls this session._ (The `diffToBlockHunks` adjacency
+  behavior was a bug caught by the accept-all E2E, not a judgment call.)
@@ -0,0 +1,95 @@
+# Session 0046.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-13T08-19 (PST)
+> Type: planning-and-executing
+> End: 2026-06-13T08-31 (PST)
+> Status: **FINALIZED.**
+> Posture: autonomous (yolo)
+
+## Launch prompt
+
+```
+/goal plan-and-execute #48 (pinning the baseline leaves the review panel fully un-annotated — zero-diff → no F3 authorship colors on unchanged blocks; proposals still show)
+
+```
+
+## Plan
+
+**#48 (story, P2)** — pinning the baseline leaves the review panel fully
+un-annotated. Anchor: leaf `story` → §4.3 R2 (no design gate). Picked
+autonomously as the next backlog item after the document-edit-flow cluster shipped
+(0043/0044/0045).
+
+When `diffBlocks(baseline, current)` yields all-`unchanged` ops (zero diff — the
+state right after a pin), the F10 on-render still author-colors every block
+(`colorByAuthor`), so the panel looks painted instead of clean. Fix: in that
+zero-diff case render clean (no author coloring, no change marks) while keeping
+`data-src` mapping (INV-36) and still injecting pending proposals (proposals are
+actions, not annotations). Narrow edge case of INV-33 — with-changes render
+unchanged; off-state unchanged; broader "authorship never colors unchanged" NOT
+in scope.
+
+Tasks:
+1. `trackChangesModel.ts` `renderReview` — detect `ops.every(unchanged)`; in that
+   case use a plain `render` (skip `colorByAuthor`) for blocks; proposal injection
+   loop unchanged.
+2. Unit: baseline==current + author spans → no `cw-by-claude`/`cw-by-human` (and
+   no `cw-add`/`cw-del`) on the body; with a pending proposal → the `cw-proposal`
+   block still renders.
+3. Host E2E: open preview, diverge + author-color, pin → panel clean (no
+   green/blue); edit again → annotations return.
+4. Content repo: one-line INV-33 clarification (zero-diff → clean on-render) in
+   `specs/coauthoring-interactive-review.md`.
+
+No new persisted artifact; no deploy pipeline (VS Code extension).
+
+## Results
+
+**#48 (story, P2) shipped to `main`** — PR
+[#52](https://git.benstull.org/benstull/vscode-cowriting-plugin/pulls/52)
+(merged), issue #48 closed. Picked autonomously after the document-edit-flow
+cluster (0043/0044/0045).
+
+- `trackChangesModel.ts` — `renderReview` gains a `pinned` `RenderOption`; when
+  pinned + zero-diff, blocks render plain (skip `colorByAuthor`).
+- `trackChangesPreview.ts` — pass `{ pinned: baseline?.reason === "pinned" }`
+  from `refresh` + `renderHtmlFor`.
+- Unit (4 new) + `s48PinClean` host E2E.
+
+**Scoped to the pin specifically** (not all zero-diff): a baseline advanced by a
+**machine-landing** (accept) is also zero-diff but keeps its authorship coloring
+(F10 INV-33) — the F10 authorship E2E caught a pure-zero-diff rule would regress
+accepted-Claude-text coloring, so the clean render is gated on `reason ===
+"pinned"`.
+
+Verification: `tsc --noEmit` clean; **218 unit** green; **74/5 host E2E** green.
+
+## Session arc
+
+1. **Pick + claim.** Stop hook required determining/executing the next milestone
+   autonomously (rolling `next`); chose #48 (leaf story, no design gate) and
+   claimed 0046. Clean `main`.
+2. **Read the issue** (detailed, gave the solution shape) + `renderReview`.
+3. **TDD.** Implemented zero-diff-clean; full suite caught the F10 authorship
+   regression (accept advances baseline → zero-diff → coloring was being cleared);
+   re-scoped to `reason === "pinned"` via a `pinned` RenderOption → all green.
+4. **Shipped** (PR #52), closed #48; spec clarification written but left in the
+   content-repo working tree (see Deferred decisions).
+
+## Deferred decisions
+
+_Autonomous-mode low-confidence calls the driver made and would have
+liked operator input on. Empty if none._
+
+- **INV-33 clarification not pushed to the content repo** (loose end, not a
+  judgment call). Wrote the zero-diff-after-pin clarification into
+  `specs/coauthoring-interactive-review.md`, but the **content repo has
+  pre-existing uncommitted state that isn't this session's**: a modified
+  `coauthoring-diff-view.md` and ~18 untracked capture-session draft files
+  (`issues/*.md`, `specs/coauthoring-document-edit-flow.md`), and local `main` is
+  **6 behind origin**. A clean rebase would require deleting the operator's
+  untracked drafts (irreversible — a STOP gate), so I **soft-reset** my spec
+  commit into the working tree rather than force it. The clarification sits as an
+  uncommitted modification alongside the operator's other content-repo drafts,
+  for the operator to reconcile/push. The #48 **code** shipped normally.
@@ -0,0 +1,86 @@
+# Session 0047.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-13T08-34 (PST)
+> Type: planning-and-executing
+> End: 2026-06-13T08-44 (PST)
+> Status: **FINALIZED.**
+> Posture: autonomous (yolo)
+
+## Launch prompt
+
+```
+/goal plan-and-execute #33 (intra-emphasis sentinel hardening — token-aware fix for the 2 characterized failure modes)
+
+```
+
+## Plan
+
+**#33 (task, P3)** — harden the F9/F10 author-coloring PUA sentinels against
+intra-emphasis markdown. Anchor: leaf `task` → §4.3 R2 (no design gate). Picked
+autonomously (next clean plan-and-executable backlog item; doesn't touch the
+content repo). Approach = the token-aware fix recommended in the issue #33 comment.
+
+Two characterized failure modes (issue #33 comment, session 0032):
+- **CASE1** — a span boundary lands strictly inside a delimiter run (`a**b**c`,
+  boundary between the two `*`): the injected sentinel splits `**` → markdown parse
+  breaks (stray `<em></em>`, raw `**` left).
+- **CASE3** — a span boundary inside an emphasis run (`**bold**`, span covers
+  `**bo`): emphasis renders but the author `<span>` and `<strong>` **misnest**
+  (`<strong>bo</span>ld</strong>`).
+
+Fix (both needed):
+1. `injectSentinels` — **clamp** any sentinel offset that lands strictly inside a
+   markdown delimiter run (`* _ ~ \``) to the run's start, so a sentinel never
+   splits a delimiter (fixes CASE1). Skip now-zero-width spans.
+2. `sentinelsToSpans` — replace the naive split/join with a **token-aware walker**
+   over the rendered HTML: emit `cw-by-*` spans only around TEXT runs, closing the
+   span before any `<tag>` and reopening after, so a span never crosses an element
+   boundary (fixes CASE3 — one span segment per text run). Strip any stray sentinel
+   left inside a tag (no PUA leakage).
+
+Pure/vscode-free/deterministic (INV-33). Non-goals: link/attribute sentinel cases,
+visual language, attribution model.
+
+Tasks: unit tests reproducing CASE1 + CASE3 (+ regression on existing
+colorByAuthor/renderReview cases) → implement clamp + walker → green; spec already
+notes the hardening (§1.7/§9). No deploy pipeline (extension).
+
+## Results
+
+**#33 (task, P3) shipped to `main`** — PR
+[#53](https://git.benstull.org/benstull/vscode-cowriting-plugin/pulls/53)
+(merged), issue #33 closed. Picked autonomously (next clean plan-and-executable
+leaf; doesn't touch the content repo).
+
+- `trackChangesModel.ts` `injectSentinels` — clamp any sentinel offset landing
+  strictly inside a delimiter run (`* _ ~ \``) to the run's start (fixes CASE1
+  parse-break); drop spans that clamp to empty.
+- `trackChangesModel.ts` `sentinelsToSpans` — token-aware walker emitting the
+  `cw-by-*` span only around text runs, split at every `<tag>` boundary (fixes
+  CASE3 misnest); strays stripped (no PUA leak).
+- 4 new unit tests (CASE1/CASE3/CASE2-regression/no-leak), real markdown-it via
+  `renderReview`.
+
+Verification: `tsc --noEmit` clean; **222 unit** green; **74/5 host E2E** green;
+the F10/authorship E2E (`cw-by-claude`/`cw-by-human`) still pass (common case
+byte-identical).
+
+## Session arc
+
+1. Stop hook required autonomously determining/executing the next milestone;
+   chose #33 (leaf task, no content-repo dependency). Claimed 0047, clean `main`.
+2. Read issue #33 + its investigation comment (2 failure modes, recommended
+   token-aware approach) + session 0032 characterization + the sentinel code.
+3. TDD: CASE1/CASE3 reproduced red → implemented clamp + walker → green; full
+   suite + E2E green; self-reviewed (one pure module).
+4. Shipped (PR #53), closed #33.
+
+## Deferred decisions
+
+_Autonomous-mode low-confidence calls the driver made and would have
+liked operator input on. Empty if none._
+
+- _No low-confidence calls this session._ (Approach was the one recommended in the
+  issue #33 comment; the fix is a pure-module robustness change with full test
+  coverage and no regression.)
@@ -0,0 +1,90 @@
+# Session 0048.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-13T08-45 (PST)
+> Type: planning-and-executing
+> End: 2026-06-13T08-58 (PST)
+> Status: **FINALIZED.**
+> Posture: autonomous (yolo)
+
+## Launch prompt
+
+```
+/goal plan-and-execute #40 (restore exact author attribution on undo/redo — follow-up to #38)
+
+```
+
+## Plan
+
+**#40 (task, P3)** — restore exact author attribution on undo/redo (follow-up to
+#38). Anchor: leaf `task` → §4.3 R2 (no design gate). Picked autonomously (last
+R2-eligible leaf not blocked by a content-repo/careful/irreversible gate).
+
+#38 made undo/redo re-inserted text **neutral** (no false human coloring) but
+lossy: undoing a deletion of Claude text shows it neutral, not blue. #40 restores
+the **exact prior** attribution.
+
+**Mechanism (engineering choice): text-keyed attribution snapshots.** Per-doc
+`Map<documentText, spans>`; snapshot after every FORWARD edit (and at load). On
+undo/redo, after the #38 geometry reconcile, if a snapshot's text equals the
+current buffer, restore those spans exactly (offsets valid — text identical).
+Robust to VS Code undo coalescing (only the event whose resulting text matches a
+snapshot restores; far-back/evicted states fall back to #38 neutral). Bounded
+history.
+
+Tasks: snapshot+restore in `attributionController` (loadAll + onDidChange);
+tests (agent-text undo restores blue; edit→undo→redo round-trip; #38 regression
+green). No deploy pipeline (extension).
+
+## Results
+
+**#40 implemented but NOT shipped — verification-blocked.** On branch
+`s40-undo-provenance` (pushed, unmerged); issue #40 **kept open**.
+
+- `attributionController.ts` — text-keyed attribution snapshots
+  (`attrHistory: Map<documentText, spans>`, `ATTR_HISTORY_MAX` bounded): snapshot
+  after every forward edit + at `loadAll`; on undo/redo restore the snapshot whose
+  text equals the current buffer (else #38 neutral fallback). Robust to undo
+  coalescing.
+- `test/e2e/suite/s40Provenance.test.ts` — agent-text-undo-restores-blue +
+  edit→undo→redo round-trip.
+- **222 unit + typecheck green.**
+
+**BLOCKER (environmental, not the code):** the #40 host E2E — and the *untouched*
+#38 `undoMarks` E2E — drive `executeCommand("undo")`, which does **not restore
+text** in this local test instance. Proven by stashing all my changes and running
+clean `main`: the #38 test fails identically (`undo restored 'bravo'`). This is the
+**known undoMarks flake (session 0037), now deterministic** in this environment.
+The usual remedy — clearing `.vscode-test/user-data` — is **permission-blocked**
+this session (`rm -rf` denied). Focusing the doc before `undo` (`showTextDocument`)
+did not help. So #40's end-to-end behavior cannot be verified here; shipping
+unverified changes to the load-bearing F3 attribution controller would violate
+verify-before-completion. Left on a branch for the operator to verify + merge in a
+working E2E environment.
+
+**Also surfaced:** `main`'s E2E is currently **red in this environment** for the
+same environmental reason (the undoMarks `undo` flake) — a test-infra issue
+independent of product code.
+
+## Session arc
+
+1. Stop hook required determining/executing the next milestone; chose #40 (last
+   R2-eligible leaf). Claimed 0048, clean `main`.
+2. Read #40 + #38 + the attribution change handler; chose text-keyed snapshots.
+3. TDD: wrote #40 E2E + implemented snapshot/restore; 222 unit green.
+4. E2E: 3 undo-driven tests failed at the `undo restored X` step. Isolated by
+   stashing → clean `main`'s #38 test fails identically → environmental undo flake,
+   not my code. Could not clear `.vscode-test` (permission-blocked).
+5. Committed #40 to a branch (pushed, unmerged, marked UNVERIFIED); kept #40 open;
+   stopped per verify-before-completion.
+
+## Deferred decisions
+
+_Autonomous-mode low-confidence calls the driver made and would have
+liked operator input on. Empty if none._
+
+- **Did not merge #40** (driver call). Alternative: merge on unit-green + reasoning
+  alone. Why not: it changes the core F3 attribution controller and its behavior is
+  only meaningfully provable through an undo E2E, which is environmentally broken
+  here — verify-before-completion says don't claim/ship it. Preserved on a branch
+  for operator verification instead. (Reversible: just merge once verified.)
@@ -0,0 +1,61 @@
+# Session 0049.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-13T09-19 (PST)
+> Type: planning-and-executing
+> End: 2026-06-13T09-24 (PST)
+> Status: **FINALIZED.**
+> Posture: autonomous (yolo)
+
+## Launch prompt
+
+```
+/goal plan-and-execute: pretest:e2e cleans stale compiled tests in out/test (stale *.test.js from other branches were running via the glob) — build hygiene follow-up to #54
+
+```
+
+## Plan
+
+Build-hygiene follow-up to #54 (noted on that issue). `tsc -p tsconfig.e2e.json`
+emits to `out/` but never removes outputs for test sources absent on the current
+branch, so stale compiled `out/test/e2e/suite/*.test.js` from other branches get
+run by the suite glob (`**/*.test.js`) — this caused real cross-branch test
+confusion in session 0048 (a removed probe + the s40 branch's tests ran on an
+unrelated branch). Fix: `pretest:e2e` cleans `out/test` before recompiling.
+
+- `package.json` — add `clean:e2e` (node `fs.rmSync('out/test', {recursive,force})`
+  — avoids shell `rm` issues) and run it between `build` and `tsc` in `pretest:e2e`.
+  Clean ONLY `out/test` (NOT `out/`, which holds the just-built esbuild bundle).
+- Verify: introduce a stale `out/test/.../zz.test.js`, run `pretest:e2e`, confirm
+  it's gone + the E2E suite is green.
+
+Trivial, ungated, verifiable; test-infra only. No deploy pipeline (extension).
+
+## Results
+
+**Shipped to `main`** — PR #56 (merged). `package.json`: new `clean:e2e`
+(`fs.rmSync('out/test',{recursive,force})`) run between `build` and `tsc` in
+`pretest:e2e`. Cleans only `out/test` (never the esbuild bundle in `out/`).
+Verified: a planted stale `out/test/.../zz.test.js` is removed by `pretest:e2e`;
+E2E green (73 passing + 1 pending [#38 undo-skip], both passes exit 0). Test-infra
+only; no product code.
+
+Properly tracked under this session (0049) — closing the protocol gap where
+#54/#55 merged after 0048 had finalized.
+
+## Session arc
+
+1. Stop hook pushed me to execute the one remaining ungated/verifiable item (the
+   `out/`-clean follow-up I'd noted on #54). Claimed 0049 to track it cleanly.
+2. Confirmed `tsc` outDir → `out/test`; esbuild bundle → `out/extension.cjs`
+   (separate), so cleaning `out/test` is safe.
+3. Implemented `clean:e2e`; verified by planting a stale compiled test (removed)
+   + full E2E green. Shipped PR #56.
+
+## Deferred decisions
+
+_Autonomous-mode low-confidence calls the driver made and would have
+liked operator input on. Empty if none._
+
+- _No low-confidence calls this session._ (Cleaning stale build output is
+  unambiguously correct; scoped to `out/test` to protect the esbuild bundle.)
@@ -0,0 +1,76 @@
+# Session 0050.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-13T17-43 (PST)
+> End: 2026-06-13T17-52 (PST)
+> Type: capture
+> Posture: autonomous (yolo)
+> Status: **FINALIZED**
+
+## Launch prompt
+
+`/wgl-capture` — "When changing documents, we should decide what happens to the
+review pane of the current document, and if the document switched to should get a
+review experience or not"
+
+## Pre-state
+
+- Branch `main`, clean, pushed (origin/main). No in-flight sessions for the app.
+- Tracker `benstull/vscode-cowriting-plugin` (host `git.benstull.org`); content
+  repo `vscode-cowriting-plugin-content` cloned, with ~18 prior uncommitted
+  capture drafts in `issues/`.
+- Prior frontier exhausted (session 0047 note): remaining backlog items gated;
+  operator directing. This is a capture session, not a build.
+
+## Arc
+
+1. **Claimed** tracked-lite session ID **0050** (`claim-session-id.sh --type
+   capture`). Resolved app + content repo (`resolve-app.py`).
+2. **Grounded** the ask by reading the code (Explore subagent over
+   `src/trackChangesPreview.ts`, `diffViewController.ts`, `proposalController.ts`):
+   confirmed the review pane is one-panel-per-doc keyed by URI
+   (`trackChangesPreview.ts:45`) with **no** `onDidChangeActiveTextEditor`
+   listener — so on document switch the pane **stays pinned** to its original doc
+   (incidental, not designed) and the switched-to doc gets no review until
+   `showTrackChangesPreview` is re-invoked. Per-doc review state (F6 baseline / F4
+   proposals / F3 attribution) auto-creates on demand.
+3. **Sized** the single ask: genuine forks (follow / pin / close; auto-review /
+   on-demand) + needs design before build → `type/feature` (R3), `priority/P2`.
+   Captures the decision-to-be-made; option space goes in Solution notes,
+   non-binding.
+4. **Drafted** `issues/review-pane-on-document-switch.md` in the content repo
+   (working tree only, INV-8), full §5 template. Scanned clean for secrets
+   (INV-3).
+5. **Operator approved** filing as-is (feature, P2) via AskUserQuestion.
+6. **Filed** → ensured labels (all pre-existing), then `capture-issues.sh` →
+   **#57** (https://git.benstull.org/benstull/vscode-cowriting-plugin/issues/57).
+
+## Cut state
+
+- Session repo `vscode-cowriting-plugin`: clean `main`, nothing committed this
+  session (capture touches only the content repo working tree). No PRs.
+- Content repo: new draft `issues/review-pane-on-document-switch.md` left
+  uncommitted for the Author to publish/discard (INV-8), alongside the prior
+  drafts already there. Not pushed by this skill.
+- Issue **#57** filed and open on the tracker.
+
+## Deferred decisions
+
+_None — single clear ask, operator approved the framing and type before filing._
+
+## Operator plate
+
+- One feature filed: **#57** (review-pane behavior on document switch, P2).
+- ~19 capture drafts now sit uncommitted in `vscode-cowriting-plugin-content` —
+  yours to commit/publish (or discard) when convenient.
+- #57 needs a design before build (R3). No active plan frontier; operator to
+  direct next.
+
+## Next session prompt
+
+`/goal` — operator to direct. Candidate: brainstorm a Solution Design for **#57**
+(review-pane behavior on document switch) — settle the two coupled rules (outgoing
+pane fate; whether the switched-to doc auto-gets a review experience) before
+building. Other open: OQ-2 (F11 graduation), #32 (scroll-sync, needs design),
+#35 (repo rename), #40 (undo provenance, P3, branch `s40-undo-provenance`
+unmerged), #54 (undo still broken in sandbox).
@@ -0,0 +1,71 @@
+# Session 0051.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-13T17-49 (PST)
+> End: 2026-06-13T17-58 (PST)
+> Type: capture
+> Posture: autonomous (yolo)
+> Status: **FINALIZED**
+
+## Launch prompt
+
+`/wgl-capture` — two asks:
+1. "Undo doesn't undo Claude's edits"
+2. "When I make a change within a block that Claude has changed that I haven't
+   approved, it doesn't show my changes"
+
+## Pre-state
+
+- Branch `main`, clean, pushed. Immediately followed capture session 0050 (#57).
+- Tracker `benstull/vscode-cowriting-plugin` (host `git.benstull.org`); content
+  repo cloned.
+
+## Arc
+
+1. **Claimed** tracked-lite session ID **0051**.
+2. **Grounded** both asks in code (Explore subagent):
+   - Undo: Claude's edits are pending proposals; `proposalController.ts:303-310`
+     shifts anchors on edit but ignores `e.reason === Undo` — proposal not
+     reconciled on undo (accepted edits hit the doc and are undoable, but the
+     pending proposal lingers).
+   - Pending-block edits: a proposal freezes its "before" (`replaced: fp.text`,
+     `proposalController.ts:81-95`); the block renders from current text but the
+     proposal's frozen view (`trackChangesModel.ts:706-722`/`:769-805`) obscures
+     live human edits in that block.
+3. **Checked duplication** against the undo cluster — #38 (wrong marks, closed),
+   #40 (attribution color, open), #54 (undo E2E infra, open). Both new asks are
+   distinct (undo failing to *reverse* the edit; live edits obscured by a pending
+   proposal). Verified `type/bug` exists on the tracker.
+4. **Drafted** both as `type/bug` (P1) in the content repo working tree
+   (`issues/undo-does-not-undo-claude-edits.md`,
+   `issues/edits-in-pending-block-not-shown.md`), each framed as user-facing pain
+   with repro steps + code diagnosis demoted to parenthetical. Scanned clean for
+   secrets (INV-3).
+5. **Operator triage:** before filing, operator said "ignore both of these
+   issues — you can delete the drafts." Both drafts **deleted**; nothing filed.
+
+## Cut state
+
+- **Nothing filed** (INV-4 — no-actionable outcome after operator triage).
+- Both draft files removed from the content repo working tree.
+- Session repo `vscode-cowriting-plugin`: clean `main`, no commits this session.
+
+## Deferred decisions
+
+_None — operator explicitly directed dropping both asks._
+
+## Operator plate
+
+- No issues filed this session. The two undo/pending-block asks were
+  considered and intentionally dropped (recorded in memory
+  `session-0051-capture-no-actionable.md` so they aren't re-raised as new).
+- Note: these remain *real* observed behaviors per the code reading — dropped on
+  operator triage, not because they were invalid. If they resurface, the prior
+  grounding is in the memory note.
+
+## Next session prompt
+
+`/goal` — operator to direct. From session 0050: candidate is brainstorming a
+Solution Design for **#57** (review-pane behavior on document switch). Other
+open: OQ-2 (F11 graduation), #32 (scroll-sync), #35 (repo rename), #40 (undo
+provenance, P3, branch `s40-undo-provenance` unmerged), #54 (undo E2E infra).
@@ -0,0 +1,73 @@
+# Session 0052.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-15T08-33 (PST)
+> End: 2026-06-15T08-41 (PST)
+> Type: capture
+> Posture: autonomous (yolo)
+> Status: **FINALIZED**
+
+## Launch prompt
+
+`/wgl-capture` — two asks (second arrived mid-session):
+1. "the review window and markdown source file should have scroll location sync'd"
+2. "the claude change recommendation block (which user will accept or decline)
+   should show the diff with the original"
+
+## Pre-state
+
+- Branch `main`, clean, pushed. Follows capture sessions 0050 (#57) and 0051
+  (no-actionable).
+- Tracker `benstull/vscode-cowriting-plugin` (host `git.benstull.org`).
+
+## Arc
+
+**Ask 1 — scroll-sync (duplicate, nothing filed):**
+1. Claimed tracked-lite session ID **0052**.
+2. Recognized scroll-sync as a known F10 follow-up; fetched **#32** — exact match
+   (*"Scroll-sync the review preview with the source editor"*, open,
+   `type/feature`, `priority/P3`, filed session 0031).
+3. Surfaced to operator; operator chose to **leave #32 at P3**. No new issue
+   (INV-4 duplicate).
+
+**Ask 2 — proposal block diff vs original (filed #58):**
+4. Grounded in code (Explore): the pending proposal block renders original +
+   proposed as **two separate full blocks** (`<del>` whole-before + `<ins>`
+   whole-after, `trackChangesModel.ts:715-716`) with **no word-level diff** — the
+   changed-block rendering already uses `wordMergedMarkdown`/`diffWords`
+   (`trackChangesModel.ts:431-438`) but `proposalBlockHtml()` doesn't.
+   `ProposalView` already carries `replaced`+`replacement`
+   (`proposalController.ts:81-95`).
+5. Sized `type/story`, P2 (reuses existing word-diff helper; distinct from #31
+   placement / #47 granularity). Drafted
+   `issues/proposal-block-shows-diff-with-original.md` (working tree only, secrets
+   scanned).
+6. Operator approved as-is → ensured labels → filed **#58**
+   (https://git.benstull.org/benstull/vscode-cowriting-plugin/issues/58).
+
+## Cut state
+
+- **#58** filed (story, P2). Scroll-sync left as existing **#32** (P3) — nothing
+  new filed for it.
+- Draft `issues/proposal-block-shows-diff-with-original.md` left uncommitted in
+  the content repo for the Author to publish (INV-8).
+- Session repo `vscode-cowriting-plugin`: clean `main`, no commits this session.
+
+## Deferred decisions
+
+_None — operator directly resolved the duplicate and approved the #58 framing._
+
+## Operator plate
+
+- One issue filed: **#58** (proposal block word-diff, P2).
+- Scroll-sync remains **#32** (P3, open, unchanged).
+- One uncommitted capture draft added to the content repo (joins the prior
+  drafts) — yours to publish or discard.
+
+## Next session prompt
+
+`/goal` — operator to direct. Standing candidate (from 0050): brainstorm a
+Solution Design for **#57** (review-pane behavior on document switch). Open work:
+#58 (proposal word-diff, P2, story), #32 (scroll-sync, P3), OQ-2 (F11
+graduation), #35 (repo rename), #40 (undo provenance, P3, branch
+`s40-undo-provenance` unmerged), #54 (undo E2E infra).
@@ -0,0 +1,66 @@
+# Session 0053.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-15T10-23 (PST)
+> End: 2026-06-15T10-29 (PST)
+> Type: capture
+> Posture: autonomous (yolo)
+> Status: **FINALIZED**
+
+## Launch prompt
+
+`/wgl-capture` — "The plugin is asking for access to other applications but works
+fine when I decline. It just shouldn't"
+
+## Pre-state
+
+- Branch `main`, clean, pushed. Follows capture sessions 0050 (#57), 0051
+  (no-actionable), 0052 (#58 + scroll-sync dup).
+- Tracker `benstull/vscode-cowriting-plugin` (host `git.benstull.org`).
+
+## Arc
+
+1. Claimed tracked-lite session ID **0053**.
+2. **Investigated** the macOS Automation/Apple-Events trigger (Explore): the
+   extension's own source has **no** AppleScript/`osascript`/automation calls; the
+   probable origin is **`@cline/sdk`/`@cline/core`** during agent activation
+   (likely local Claude Code hub-discovery), which falls back gracefully when
+   denied — matching "works fine when declined." Framed as a lead, not a settled
+   root cause.
+3. **Sized** `type/bug` (declared defect: unwanted OS permission request).
+   Drafted `issues/spurious-automation-permission-prompt.md` (working tree only,
+   secrets scanned) with repro steps + expected/actual + the upstream caveat.
+4. **Operator set priority P1** (trust/launch-optics over the harmless functional
+   impact); updated the WSJF line accordingly.
+5. **Filed #59.** capture-issues.sh `--type` rejects `bug` (known #124 taxonomy
+   gap), so filed with `--type task` then swapped the label to `type/bug` via the
+   Gitea API (deleted type/task, added type/bug) — verified `['priority/P1',
+   'type/bug']` (INV-2 holds).
+   → https://git.benstull.org/benstull/vscode-cowriting-plugin/issues/59
+
+## Cut state
+
+- **#59** filed (bug, P1), label corrected to exactly `type/bug`.
+- Draft `issues/spurious-automation-permission-prompt.md` left uncommitted in the
+  content repo for the Author to publish (INV-8).
+- Session repo `vscode-cowriting-plugin`: clean `main`, no commits this session.
+
+## Deferred decisions
+
+_None — operator set the priority directly and approved the framing._
+
+## Operator plate
+
+- One issue filed: **#59** (spurious macOS automation prompt, bug, **P1**).
+- The cause is likely upstream in `@cline/sdk` — the fix may be a config/flag or
+  dependency update, not necessarily this repo's code.
+- Reminder: bug capture needs the file-as-task-then-relabel workaround until
+  plugin #124 (capture can't type bugs) is fixed.
+
+## Next session prompt
+
+`/goal` — operator to direct. Standing candidate (from 0050): brainstorm a
+Solution Design for **#57** (review-pane behavior on document switch). Open work:
+**#59** (automation prompt, bug P1 — may be upstream), #58 (proposal word-diff,
+story P2), #32 (scroll-sync, P3), OQ-2 (F11 graduation), #35 (repo rename), #40
+(undo provenance, P3), #54 (undo E2E infra).
@@ -0,0 +1,66 @@
+# Session 0054.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-15T10-42 (PST)
+> End: 2026-06-15T10-48 (PST)
+> Type: capture
+> Posture: autonomous (yolo)
+> Status: **FINALIZED**
+
+## Launch prompt
+
+`/wgl-capture` — "See the Claude output/progress in the 'asking Claude…' status"
+
+## Pre-state
+
+- Branch `main`, clean, pushed. Follows capture sessions 0050 (#57), 0051
+  (no-actionable), 0052 (#58 + scroll-sync dup), 0053 (#59).
+- Tracker `benstull/vscode-cowriting-plugin` (host `git.benstull.org`).
+
+## Arc
+
+1. Claimed tracked-lite session ID **0054**.
+2. **Grounded** (Explore): the "Cowriting: asking Claude…" status is an opaque
+   `withProgress` notification (`extension.ts:232-233`,
+   `trackChangesPreview.ts:232-233`) awaiting `agent.run()` as one black-box
+   promise (`liveTurn.ts:52`). Key enabler: `@cline/sdk` already emits streaming
+   events (`assistant-text-delta`, `tool-started/updated/finished`,
+   `usage-updated`) via `agent.subscribe()` / an `onEvent` hook, but the extension
+   constructs the Agent with **no hooks** (`liveTurn.ts:47-51`) and discards it.
+3. **Sized** `type/feature` — implementable by subscribing to existing events, but
+   a real design fork on the surface (notification text vs. OutputChannel vs.
+   status bar vs. webview relay) and content (text/tool/usage/reasoning) → design
+   first (R3). Drafted `issues/show-live-claude-progress.md` (working tree only,
+   secrets scanned).
+4. **Operator set priority** — chose "adjust priority" → **P1** (opaque wait hurts
+   every turn). Updated WSJF line.
+5. **Filed #60** (`type/feature`, P1).
+   → https://git.benstull.org/benstull/vscode-cowriting-plugin/issues/60
+
+## Cut state
+
+- **#60** filed (feature, P1).
+- Draft `issues/show-live-claude-progress.md` left uncommitted in the content repo
+  for the Author to publish (INV-8).
+- Session repo `vscode-cowriting-plugin`: clean `main`, no commits this session.
+
+## Deferred decisions
+
+_None — operator set the priority directly and approved the framing._
+
+## Operator plate
+
+- One issue filed: **#60** (live Claude progress, feature, **P1**).
+- Implementation is largely "subscribe to SDK events already available + pick the
+  progress surface" — needs a small design first (R3).
+- One uncommitted capture draft added to the content repo — yours to publish or
+  discard.
+
+## Next session prompt
+
+`/goal` — operator to direct. Capture run 0050–0054 filed #57, #58, #59, #60
+(scroll-sync = existing #32). Two P1s now open (**#59** automation prompt,
+**#60** live progress) plus standing #57 (review-pane on doc switch, design
+candidate). Other open: #58 (proposal word-diff, P2), #32 (scroll-sync P3), OQ-2
+(F11 graduation), #35 (repo rename), #40 (undo provenance P3), #54 (undo E2E
+infra).
@@ -0,0 +1,105 @@
+# Session 0055.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-22T23-19 (PST)
+> End: 2026-06-26T04-23 (PST)
+> Type: brainstorming
+> Posture: autonomous (yolo)
+> Status: **FINALIZED**
+
+## Launch prompt
+
+`/wgl-brainstorming highest priority ticket`
+
+## Plan
+
+Brainstorming session: explore the highest-priority open ticket and write its
+Solution-Design spec. Highest-priority that fits a brainstorming session (spec
+output; a feature needs a design per §4.3 R3) was **#60 — Show Claude's live
+output/progress during the "asking Claude…" status** (P1 feature). #59 (P1 bug)
+is coding-session work, not brainstorming.
+
+Output: `specs/coauthoring-live-progress.md` (Solution Design), submitted graduated
+at finalize to the content repo's `specs/` collection.
+
+## Pre-state
+
+- Clean, pushed `main` (`24e329e`). No in-flight sessions (next ID 0055).
+- Two P1 tickets tied for highest priority: **#59** (P1 bug — spurious macOS
+  "control other applications" prompt) and **#60** (P1 feature — live turn
+  progress). Brainstorming produces a spec; a feature needs a design (§4.3 R3),
+  so #60 was the fit. #59 (bug) → a coding session.
+- Highest INV in use across the content-repo specs: **INV-42**.
+
+## Session arc (turn-by-turn)
+
+1. **Gate + classify.** Launch prompt classified as **brainstorming**; ran
+   `wgl-brainstorming`.
+2. **Claim (peek → gate → claim).** `claim-session-id.sh --dry-run` → no in-flight
+   sessions, next ID **0055**. Claimed `--type brainstorming` (placeholder at
+   `16cca30`). Verified clean pushed `main` baseline.
+3. **Orient.** Surveyed the open tracker. Picked **#60** (P1 feature, design
+   needed) over #59 (P1 bug, coding-session work). Read #60 in full.
+4. **Ground in code (read-only).** Read `liveTurn.ts` (`runEditTurn` →
+   `agent.run()` as one opaque promise; module is deliberately vscode-free), both
+   call sites (`extension.ts:232` editSelection, `trackChangesPreview.ts:232`
+   preview `askClaude` — both wrap the turn in `withProgress("asking Claude…")`),
+   the injectable `editTurn` seam (trackChangesPreview.ts:56, for host-E2E
+   stubbing), and the `@cline/sdk` event API: `@cline/agents` `Agent` exposes
+   `subscribe(listener) → unsubscribe`, `run()`, `abort()`; `AgentRuntimeEvent`
+   (in `@cline/shared/dist/agent.d.ts`) carries `assistant-text-delta`
+   (+`accumulatedText`), `tool-started|updated|finished`, `usage-updated`,
+   lifecycle events. The enabler the issue cites is real.
+5. **Brainstorm forks (superpowers:brainstorming).** Skipped the visual companion
+   (choices are conceptual/native-UI, not visual mockups). Presented three forks
+   via AskUserQuestion; operator picked all three recommendations:
+   - **Surface** → notification activity-line + token count **and** a shared
+     `"Cowriting"` OutputChannel streaming full assistant text. (Rejected: webview
+     relay can't cover the editSelection path; status-bar-only; OutputChannel-
+     primary.)
+   - **Content** → activity + token count (reasoning text not surfaced).
+   - **Cancellation** → `cancellable` notification → `agent.abort()` →
+     "cancelled", proposes nothing (reflect + cancel button).
+6. **Design presented + approved.** Three-unit architecture (pure
+   `turnProgress.ts` reducer → `runEditTurn` extended with `onProgress` +
+   `AbortSignal` → both call sites relay via a shared `liveProgressUi`); layering
+   rule keeps `liveTurn.ts`/`turnProgress.ts` vscode-free; INV-43..47. Two
+   sub-decisions (OutputChannel auto-reveal gated by a new setting; append-not-
+   clear) confirmed. Operator: "design looks right."
+7. **Spec written + self-review.** `specs/coauthoring-live-progress.md` authored
+   in house Solution-Design format. Self-review: no placeholders, INV-43..47 (no
+   collision with ≤42; INV-8/21/39 are correct citations), internally consistent,
+   single-feature scope.
+8. **Operator: "Implement it."** Per the pipeline, that's the brainstorm→coding
+   handoff (one SPEC = one plan = one execution, §4.3). Finalized this
+   brainstorming session (below); a fresh `wgl-planning-and-executing` session
+   plans + builds #60.
+
+## Cut state (at finalize)
+
+- **Spec graduated.** `submit-spec.sh --status graduated` →
+  `vscode-cowriting-plugin-content` `specs/coauthoring-live-progress.md` at
+  `3bd1ae5`.
+- **Code repo:** clean `main`; no code changes this session (brainstorming).
+  A local stray `specs/coauthoring-live-progress.md` remains untracked in the code
+  repo (an `rm` cleanup was permission-denied; harmless — canonical copy is in the
+  content repo).
+- **Memory:** added `session-0055-live-progress-spec-graduated.md` + index line.
+
+## Next-session prompt
+
+```
+/wgl-planning-and-executing implement #60 (live turn progress) from coauthoring-live-progress.md
+```
+
+## Deferred decisions
+
+- **OutputChannel auto-reveal** (fires every Claude turn): decided to auto-`show(true)`
+  (preserveFocus) on the *first* text delta, gated by a new setting
+  `cowriting.liveProgress.revealOutput` (default `true`). Operator confirmed the
+  design including this.
+- **OutputChannel history:** decided to *append* per-turn (with a header) rather
+  than clear, so it doubles as a debug log. Operator confirmed.
+- **Stray local spec file:** `rm -rf specs/` was permission-denied; left the
+  untracked local copy in the code repo (cosmetic; content repo holds the
+  canonical copy).
@@ -0,0 +1,106 @@
+# Session 0056.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-26T04-24 (PST)
+> End: 2026-06-26T04-54 (PST)
+> Type: planning-and-executing
+> Posture: autonomous (yolo)
+> Status: **FINALIZED**
+
+## Launch prompt
+
+`/wgl-planning-and-executing implement #60 (live turn progress) from coauthoring-live-progress.md`
+
+(Continuation of brainstorming session 0055, which graduated the spec; the
+operator said "implement it" → this coding session.)
+
+## Plan
+
+Plan + execute **#60** (P1 feature — live turn progress) from the graduated
+Solution Design `coauthoring-live-progress.md`. Four cuts: (1) pure
+`turnProgress.ts` reducer + tests; (2) `runEditTurn` + `onProgress`/`AbortSignal`;
+(3) `liveProgressUi` + setting + both call sites; (4) host-E2E + smoke.
+Branch `s60-live-progress` → PR → main.
+
+## Pre-state
+
+- Clean `main` at session start (fast-forwarded 13 commits — the sessions/ repo
+  is the code repo itself). Baseline: 222 unit green.
+- #60 has a graduated Solution Design → §4.3 R2(b) satisfied.
+
+## Session arc
+
+1. **Claim + plan.** Claimed 0056 (no in-flight). Wrote the implementation plan
+   with `superpowers:writing-plans` → `docs/superpowers/plans/2026-06-26-live-turn-progress.md`
+   (self-reviewed: full spec coverage, no placeholders). Grounded in the exact SDK
+   types (`AgentToolCallPart.toolName`, `AgentUsage.inputTokens+outputTokens`), the
+   vitest/host-E2E patterns, and the injectable `editTurn` seam. Branch
+   `s60-live-progress`.
+2. **Task 1 — pure reducer.** `src/turnProgress.ts` + `test/turnProgress.test.ts`
+   (13 cases). Fixed the type-only event import: `AgentRuntimeEvent` is exported
+   from **`@cline/shared`**, not `@cline/sdk` (which doesn't re-export it) — caught
+   by typecheck; amended the commit.
+3. **Task 2 — `runEditTurn`.** Added `RunEditTurnOptions {onProgress, signal}`;
+   subscribes to the agent, folds events through the reducer, wires `signal` →
+   `agent.abort()`; `finally` unsubscribes. Mocked-agent unit tests (vitest
+   `doMock` intercepts the dynamic import).
+4. **Task 3 — host relay.** `src/liveProgressUi.ts` (notification line +
+   "Cowriting: Claude" OutputChannel, reveal gated by setting) + `package.json`
+   `contributes.configuration` `cowriting.liveProgress.revealOutput`.
+5. **Tasks 4+5 — both call sites.** `editSelection` (extension.ts) and preview
+   `askClaude`/`runEditAndPropose` (trackChangesPreview.ts) made cancellable,
+   relaying via the shared `liveProgressUi`; widened the `EditTurn` seam to accept
+   `opts?` (back-compat). Added `liveProgressUi` to `CowritingApi`.
+6. **Task 6 — host E2E.** `test/e2e/suite/liveProgress.test.ts`: progress is
+   additive (INV-44), aborted turn proposes nothing (INV-47).
+7. **Task 7 — verify + smoke.** typecheck + 237 unit + build green; enhanced
+   `scripts/smoke-live-turn.mjs` to log progress.
+8. **Subagent review.** Found a **Medium** (a pre-aborted `AbortSignal` was a
+   no-op — `agent.abort()` runs before the SDK's AbortController exists, so the
+   turn ran to completion) + a **Low** (a throwing `onProgress` relay could fail
+   the turn). Fixed both: short-circuit (throw) before `run()` on a pre-aborted
+   signal; wrap the relay in try/catch. Strengthened the unit + E2E to prove it.
+9. **Ship.** Pushed; opened PR #61; operator approved merge despite the
+   concurrency (below). Squash-merged `644885c`; issue #60 auto-closed.
+
+## ⚠️ Concurrency event (mid-session)
+
+Partway through, the shared working tree gained **uncommitted changes this session
+did NOT make** — a `cowriting.edit` / `routeEdit` reach refactor across
+`src/workspacePath.ts`, `test/workspacePath.test.ts`, and the menu/`routeEdit`
+parts of `package.json` + `src/extension.ts`. It broke 3 E2E (`f12Reach` ×2,
+`f11Toolbar` ×1) that still expect the old `editSelection`/`editDocument` command
+surface. Handling: committed **nothing** of it, discarded **nothing**; verified
+#60 green in isolation by stashing it (by path, reversible) — 237 unit + both #60
+E2E pass, only the documented `F10 #38` undo-sandbox flake remains — then popped
+the stash to restore it. Surfaced to the operator, who chose to merge #60 (its 7
+commits are isolated; the PR diff contains none of the foreign WIP).
+
+## Cut state (at finalize)
+
+- **#60 merged** to main (PR #61 squash `644885c`); issue #60 closed.
+- **Plan archived:** `submit-plan.sh` → content repo `plans/2026-06-26-live-turn-progress.md` (`22ec57e`).
+- **Foreign refactor WIP:** still uncommitted in the working tree, untouched —
+  for whoever is doing it to continue/commit (it rebases onto the new main).
+- **Local tree:** left on branch `s60-live-progress` (merged) DELIBERATELY — a
+  `checkout main`/`pull` would have disturbed the foreign uncommitted WIP. Local
+  main not synced; origin/main has #60. Operator should reconcile the local tree
+  + the in-flight refactor.
+- **Memory:** added `session-0056-60-live-progress-shipped.md` + index line.
+
+## Next-session prompt
+
+```
+/wgl-planning-and-executing reconcile the local tree (sync main; land/branch the in-flight cowriting.edit/routeEdit refactor), then pick the next item (open: #59 P1 bug, #57, #58, #32, #35, #40, OQ-2)
+```
+
+## Deferred decisions
+
+- **Left local on `s60-live-progress` (did not sync main):** to avoid disturbing
+  the concurrent uncommitted refactor in the shared tree. Low confidence this is
+  the tidiest end state, but it is the safest for the foreign WIP. Operator to
+  reconcile.
+- **Token field for the activity line:** used `inputTokens + outputTokens` from
+  `AgentUsage` (no `totalTokens` field exists). Cache tokens excluded.
+- **OutputChannel append-not-clear + auto-reveal gating:** per spec §3.5; carried
+  from session 0055's confirmed sub-decisions.
@@ -0,0 +1,111 @@
+# Session 0057.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-26T04-35 (PST)
+> End: 2026-06-26T06-55 (PST)
+> Type: executing-plans
+> Posture: yolo
+> Claude-Session: 3d66a467-8026-472d-9693-52a37939d493
+> Goal: Operator-feedback UX fixes on the Ask-Claude-to-Edit experience (broken
+> keybinding, missing/duplicated shortcuts, edit-box placement) — then iterate the
+> edit-input UX to its final form.
+> Outcome: Shipped two PRs to vscode-cowriting-plugin main (#62, #65); fixed and
+> shipped a dev-plugin bug (#134 / PR #135, v0.60.0) surfaced mid-session.
+
+## Plan
+
+Operator opened with three asks on the Ask-Claude UX: the live/review-panel
+shortcut didn't work, there was no shortcut for "Ask Claude to Edit
+Selection/Document" (wanted one combined shortcut), and the edit box appeared in
+a disliked spot (near the command palette) — with a request to recommend a
+VS-Code-idiomatic placement. Routed as an executing-plans (yolo) session.
+
+The session became an extended live iterate-and-verify loop on the edit-input UX,
+driven by the operator testing each build in an Extension Development Host.
+
+## Pre-session state
+
+- On branch `s60-live-progress`; `#60` (live turn progress) work in flight.
+- A **concurrent session (0056)** was live in the SAME working checkout, actively
+  committing/stashing `#60` — which clobbered this session's uncommitted edits
+  twice before isolation.
+
+## Turn-by-turn arc
+
+1. **Phase A (keybindings + command unify).** Added one `cowriting.edit` command
+   that routes selection-vs-document at runtime via a pure `routeEdit` helper
+   (unit-tested); hid the two underlying commands from the palette; collapsed the
+   split menus to one entry. Added `⌘⌥E`/`Ctrl+Alt+E` (edit) and `mac` variants
+   for `⌘⌥R`/`Ctrl+Alt+R` (review panel) — the missing `mac` variant was why the
+   panel shortcut "didn't work" (Option-key combos are unreliable on macOS).
+
+2. **Concurrent-checkout incident + recovery.** Discovered session 0056 was
+   clobbering the shared tree (git stash/pop/commit out from under this session).
+   Exported the edits as a durable patch, created an **isolated git worktree**, and
+   moved all work there. (§5.4 lesson — should have isolated up front.)
+
+3. **Input-box placement (design Q).** Recommended + implemented an inline input
+   via the Comments API; then iterated heavily per operator feedback:
+   inline-at-selection → top-of-document → **finally a multi-line webview in a
+   split pane below the document for BOTH scopes**. Each iteration was built and
+   verified live in the EDH.
+
+4. **Dev-plugin bug #134 (the root cause of the clobber).** Filed plugin feedback
+   #134 (two sessions can share one working tree — isolation was prose, not a
+   deterministic guard), then implemented the fix: `claim-session-id.sh` now stamps
+   `> Checkout:` and **refuses a claim when another live session occupies the same
+   checkout** (self-session excepted; `WGL_ALLOW_SHARED_CHECKOUT=1` override).
+   Shipped as PR #135 / v0.60.0 and made live in the running install.
+
+5. **Final edit-input form (PR #65).** Unified both scopes on the split-below
+   multi-line webview (`editInstructionInput.ts` / `promptEditInstruction`):
+   auto-focus, ⌘↵ send, Esc cancel-with-confirm-if-text, split collapses + focus
+   restored to the doc on close (no Output/Debug panel popping), selection stays
+   highlighted above. **Deleted the entire inline-comment mechanism** (`inlineAsk.ts`,
+   `cowriting.askClaude.submit/cancel`, their menus, the Escape keybinding) — net
+   deletion. Verified interactively, merged.
+
+## Cut state (end of session)
+
+| Repo | Change | Ref |
+| --- | --- | --- |
+| vscode-cowriting-plugin | Phase A: unify `cowriting.edit` + `routeEdit` + mac keybindings + inline Comments-API input | PR #62 → main (squash `9432300`) |
+| vscode-cowriting-plugin | Final: split-below multi-line webview for both scopes; remove inline-comment mechanism | PR #65 → main (squash `7e42d11`) |
+| wiggleverse-dev-claude-plugin | Deterministic shared-checkout guard in `claim-session-id.sh` + Step 6b docs; v0.60.0 | PR #135 → main (merge `ca383c4`); installed live |
+| wiggleverse-dev-claude-plugin (tracker) | Filed + closed feedback #134 (shared-checkout hazard) | issue #134 `resolution:done` |
+
+- All worktrees (`s60-edit-ux`, `edit-top-anchor`, dev-plugin `issue134`) removed;
+  branches deleted; all temporary EDH windows closed.
+- 242 unit + typecheck (src + e2e) + build green on the final tip. E2E electron
+  suite not run (known env-broken locally); webview UX verified interactively.
+- Local `main` is 3 behind `origin/main` (sessions 0058/0059 + PR #65 merge) —
+  cosmetic; a future session pulls.
+- Untracked strays `docs/superpowers/plans/2026-06-26-live-turn-progress.md` and
+  `specs/` predate this session (from the `#60` line / a known `specs/` stray);
+  left untouched — not this session's to land.
+
+## Deferred decisions
+
+None logged. No silent low-confidence calls — every judgment call (worktree
+recovery, each UX fork, cancel/placement/size trade-offs) was surfaced live and
+decided by the operator or flagged as a hard VS Code API limit.
+
+## What lands on the operator's plate
+
+- Nothing blocking. The edit-input UX line is complete and shipped.
+- The dev-plugin shared-checkout guard (#134/v0.60.0) is **live for new
+  sessions** (this session's running install was updated; new sessions get it on
+  start).
+- Known VS Code API limits captured in memory ([[ask-claude-edit-input-ux]]): no
+  multi-line input at the command-palette location; no API to set an editor split
+  ratio; no API to read bottom-panel visibility.
+
+## Prompt the operator can paste into the next session
+
+This session's thread (Ask-Claude edit UX) is **complete** — there is no single
+forced next step. Open backlog items the operator may choose from (each its own
+session): `#32` scroll-sync (feature, needs design), `#35` repo rename, `#40` undo
+provenance (P3), F11 spec graduation (OQ-2), content-repo draft reconciliation.
+
+No `Next /goal:` is recorded — the next move is an operator pick from the backlog
+above.
@@ -0,0 +1,85 @@
+# Session 0058.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-26T05-12 (PST)
+> End: 2026-06-26T05-58 (PST)
+> Type: brainstorming
+> Posture: autonomous (yolo)
+> Status: **FINALIZED**
+
+## Launch prompt
+
+Operator opened by asking what was in the Gitea backlog, then directed a
+brainstorming session for a new feature: *"We should also show the proposed change
+in the Markdown file in the same way. The proposed change in the markdown file
+should be human-editable, show the diff just like it would after the change is
+completed in both the markdown file and the webview as the change is made. The
+diff in both the markdown file and preview webview should be exactly as they will
+be shown if the edit is accepted. Create a new working tree in a directory other
+than main for this work."* — extended mid-session with: *"replace the checkbox
+icon and 'x' icon on the proposed change as 'Accept' and 'Reject'. Each should
+have an arrow next to them for a dropdown that is 'Accept All' or 'Reject All'.
+One should be able to accept, reject, accept all, or reject all from either the
+markdown file or the web preview file."*
+
+## Pre-state
+
+- Branch `main`, clean/pushed baseline (after FF). Latest shipped work: #60 live
+  turn progress (session 0056, PR #61). Session 0057 (a `cowriting.edit`/`routeEdit`
+  refactor) still **in-flight** sharing the checkout.
+- Backlog at session start (8 open): #59 (P1 bug, Apple-Events prompt), #54/#57/#58
+  (P2), #32/#35/#40 (P3), #62 (untriaged).
+
+## Turn-by-turn arc
+
+1. **Backlog query** — listed the 8 open Gitea issues for `benstull/vscode-cowriting-plugin`
+   by priority; flagged #62 as the only untriaged item.
+2. **Session routing** — operator's feature request was feature-shaped *and*
+   reversed a locked invariant (F10/INV-32 "clean editor"), so the session type
+   was ambiguous. Asked; operator chose **brainstorm a spec first**. Routed via
+   `wgl-brainstorming`; claimed session **0058**.
+3. **Isolation** — session 0057 still in-flight on the shared checkout → created an
+   isolated worktree `vscode-cowriting-plugin-s58` on branch
+   `s58-inline-editor-diff` off `origin/main` (§5.4).
+4. **Exploration** — dispatched an Explore agent that mapped five areas: webview
+   render engine (`trackChangesModel.ts`), F4 proposal model + seam, the F10
+   clean-editor decision (zero editor decorations remain), controllers/wiring, and
+   commands/toolbar. Key constraint surfaced: **VS Code text editors cannot host
+   editable phantom text** — anything editable in the editor is buffer content;
+   decorations only style existing text or inject *non-editable* before/after
+   content.
+5. **Four design forks (AskUserQuestion, with ASCII previews):**
+   - Editor model → **optimistic apply + decorations**.
+   - Timing → **on proposal** (not live token-stream into the editor).
+   - Editor affordance → **CodeLens `Accept ▾ / Reject ▾` + QuickPick dropdown**.
+   - (Operator-added) Control parity → Accept/Reject/Accept-all/Reject-all from
+     both surfaces; ✓/✗ superseded.
+6. **Design presented** section-by-section; operator approved with "keep going"
+   (nodded the two flagged sub-decisions: dedicated `EditorProposalController`;
+   allow save of pending text).
+7. **Anchor + spec** — filed Feature **#64** (`type/feature`, `priority/P2`) since
+   no tracker anchor existed; wrote `specs/coauthoring-inline-editor-diff.md`
+   (INV-48..54, reversing INV-10/INV-32). Self-review clean. Operator **approved**.
+
+## Cut state
+
+- Spec **graduated** → content repo `vscode-cowriting-plugin-content` at `80d5d1a`
+  (`specs/coauthoring-inline-editor-diff.md`).
+- Feature **#64** filed + labelled.
+- Worktree `vscode-cowriting-plugin-s58` (branch `s58-inline-editor-diff`) stands
+  ready for the downstream planning-and-executing session; no code committed this
+  session (brainstorming output is the spec only).
+- Memory updated: `f12-inline-editor-diff-spec-graduated.md` + MEMORY.md index.
+
+## Deferred decisions
+
+No autonomous low-confidence calls — every fork was decided live with the operator
+via AskUserQuestion, and the spec was operator-approved before submission. (The
+two design sub-decisions — dedicated controller, allow-save-of-pending — were
+flagged in the presented design and nodded by the operator.)
+
+## Next-session prompt
+
+```
+/goal Plan-and-execute #64 (F12 inline editable proposed-change diff in the Markdown editor) from specs/coauthoring-inline-editor-diff.md, in worktree vscode-cowriting-plugin-s58 (branch s58-inline-editor-diff); follow the spec's 5-slice Delivery Plan
+```
@@ -0,0 +1,101 @@
+# Session 0059.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-26T06-01 (PST)
+> End: 2026-06-26T08-32 (PST)
+> Type: planning-and-executing
+> Posture: autonomous (yolo)
+> Status: **FINALIZED**
+
+## Launch prompt
+
+Plan-and-execute #64 (F12 inline editable proposed-change diff in the Markdown
+editor) from the graduated spec `specs/coauthoring-inline-editor-diff.md`, in
+isolated worktree `vscode-cowriting-plugin-s58` (branch `s58-inline-editor-diff`),
+opened straight off brainstorming session 0058.
+
+## Plan
+
+Implement #64 per spec: optimistic-apply the proposed change into the editor
+buffer (editable, tinted insertions + struck-red deletion hints), accept =
+finalize-in-place / reject = revert-in-place, `Accept ▾`/`Reject ▾` (+ Accept-all
+/Reject-all) from both editor (CodeLens+QuickPick) and webview. Reverses
+INV-10/INV-32; INV-48..54. Implementation plan:
+`docs/superpowers/plans/2026-06-26-inline-editor-diff.md`.
+
+## Outcome — SHIPPED
+
+#64 (F12) **shipped to main** — PR #66 squash `7b98249`, issue auto-closed.
+Worktree built, subagent-driven (fresh implementer per task + two-stage review),
+adversarial final review, then merged through the concurrent #65 landing.
+
+## Turn-by-turn arc
+
+1. **Init** — claimed session 0059 (0057 still in-flight; isolated worktree made
+   it §5.4-safe). Synced worktree, `npm ci`, green baseline (242 unit).
+2. **Plan** — read the whole subsystem (proposalController, trackChangesModel,
+   attributionController, trackChangesPreview, preview.ts/css, extension.ts,
+   package.json, E2E harness); wrote an 8-task TDD plan. Surfaced a **spec
+   refinement**: optimistic apply must re-anchor the proposal to the applied text
+   + store `Proposal.original` (F4's `fp.text` is the original, gone once applied).
+3. **Execute (subagent-driven, sonnet implementers):**
+   - T1 `Proposal.original` (model) — `bea9fd5`.
+   - T2 `setProposalApplied` re-anchor helper — `92ff4dd`.
+   - T3 seam `landBaseline` opt + `signalLanded` — `6d54963`. **Surfaced the
+     host-E2E was blocked by a macOS UNIX-socket-length `EINVAL`** (the long
+     worktree path). Fixed by pinning `--user-data-dir` to a short `/tmp` path
+     (`930b4ab`) → host-E2E now launches from a worktree.
+   - T4 `optimisticApply`/`finalizeInPlace`/`revertInPlace`/`rejectAll` —
+     `ddbbb7a` (subagent caught a real `keyOf` vs `uri.toString()` test bug).
+   - T5 `EditorProposalController` + pure `decorationPlan` — `8c6e782`+`e599284`
+     (subagent added a setTimeout(0) apply-debounce for the batch race; updated 5
+     existing E2E for the INV-10 reversal — verified legitimate, not weakened).
+   - T6 render-once INV-50 reconciliation — `6529332`; I added a coloring-
+     alignment fix (`landedSpans` via `toLanded`) the subagent's note flagged —
+     `3805323`.
+   - T7 webview `Accept ▾`/`Reject ▾` + dropdown + `rejectAllProposals` — `f4594da`.
+4. **Adversarial final review (opus subagent)** — found a **CRITICAL** (revert
+   target clobbered on save+reload, in-memory `applied` empty post-reload) and a
+   MINOR (toolbar summary double-counts). Both fixed + a reload-safety regression
+   E2E — `6560894`. Verdict invariants INV-48..54 upheld; decorationPlan alignment
+   verified correct.
+5. **Concurrency** — session 0057 had merged **#65** (unify Ask-Claude on a
+   split-below webview; removed `InlineAskController`) to main mid-session. Merged
+   origin/main into the branch, resolved the one `extension.ts` conflict (drop
+   inlineAsk, keep `EditorProposalController`); suite green; squash-merged PR #66.
+6. **Hygiene** — a `git add -A` during the merge accidentally tracked the spec
+   copy in the code repo; removed it (PR #67 `844a278`) — specs are canonical in
+   the content repo. Plan archived (`8c14f54`); spec patched to **v0.1.1** with the
+   re-anchor refinement and submitted (`43913dc`).
+
+## Cut state
+
+- **main @ `844a278`** (clean, pushed). #64 closed.
+- Tests: **249 unit + 87/5 host-E2E green**, typecheck clean.
+- Content repo: spec v0.1.1 (`43913dc`) + plan (`8c14f54`).
+- Worktree removed.
+
+## Deferred decisions
+
+- **Re-anchor on optimistic apply (spec refinement, RESOLVED into v0.1.1).** Added
+  `Proposal.original` + re-fingerprint to the applied text; captured once →
+  reload-safe. Folded into the spec.
+- **`finalizeInPlace` has no orphan re-resolve guard** (CodeLens Accept), unlike
+  the webview `acceptById`. Intentional — spec §2.2 says Accept keeps the human's
+  in-place edits (which would orphan the exact-text fp) — but it's asymmetric. Left
+  as-is; noted as a follow-up nit.
+- **Per-proposal webview carets always show Accept-all/Reject-all** even at 1
+  pending (the toolbar button correctly hides <2). Cosmetic; not fixed.
+
+## Not done
+
+- **Manual smoke (interactive F5)** — the live editor diff, CodeLens dropdown,
+  both-surface parity, and save-while-pending need an Extension Development Host
+  run (operator-side). No flotilla/PPE for this app — the PR-merge to main is the
+  ship.
+
+## Next-session prompt
+
+```
+/goal Manually smoke-test #64 (F12 inline editor diff) in an Extension Development Host (F5) — verify the live editable diff + deletion hints, the Accept ▾/Reject ▾ CodeLens dropdown, editor↔webview parity, and save-while-pending — then start the next backlog item; #59 (P1, macOS Apple-Events prompt, likely @cline/sdk) is highest priority
+```
@@ -0,0 +1,124 @@
+# Session 0060.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-26T12-52 (PST)
+> End: 2026-06-26T13-14 (PST)
+> Type: planning-and-executing
+> Posture: autonomous (yolo)
+> Status: **FINALIZED**
+
+## Launch prompt
+
+`/goal next` — resume the stored next-goal from session 0059's finalize:
+"Manually smoke-test #64 in an Extension Development Host (F5), then pick the
+next backlog item — #59 (P1 Apple-Events prompt) is highest priority."
+
+The #64 F5 smoke is an interactive operator gesture (flagged, not performed by
+the session). The autonomous substantive work was **fixing #59** — a standalone
+leaf `type/bug` (P1), eligible for plan-and-execute under §4.3 R2(b) with no
+design required.
+
+## Plan
+
+Fix #59: the plugin triggers a spurious macOS "control other applications"
+(Apple Events / Automation TCC) permission prompt; works fine when declined.
+Root-cause via systematic-debugging, fix via TDD, ship through branch → PR →
+merge to main.
+
+## Turn-by-turn arc
+
+1. **Session init.** Classified `/goal next` → planning-and-executing; claimed
+   session 0060 (`claim-session-id.sh`, no other sessions in flight). Baseline:
+   pulled `main` (was 1 behind), clean apart from two pre-existing stray
+   untracked files (`specs/coauthoring-live-progress.md`,
+   `docs/superpowers/plans/2026-06-26-live-turn-progress.md` — canonical copies
+   live in the content repo; left untouched).
+
+2. **Read #59** (Gitea API). P1, type/bug. Lead in the issue: not the extension's
+   own code; probable origin `@cline/sdk` agent activation / Claude-Code
+   hub-discovery, which falls back gracefully when denied.
+
+3. **Root-cause investigation (systematic-debugging Phase 1–2).**
+   - The extension's own code and the **entire JS dependency tree** contain no
+     osascript/AppleScript (only the build-only `vite`). The Apple Event is sent
+     by a spawned subprocess.
+   - Traced the spawn chain: `src/liveTurn.ts` `new sdk.Agent({providerId:
+     "claude-code"})` → `@cline/sdk` → `@cline/llms` `createClaudeCodeProviderModule`
+     → `ai-sdk-provider-claude-code` `createClaudeCode(options)` → `@anthropic-ai/
+     claude-agent-sdk` `query()` → **spawns the bundled `claude` CLI**
+     (`@anthropic-ai/claude-agent-sdk-darwin-arm64/claude`) headless
+     (`--print --output-format stream-json`).
+   - `runEditTurn` (liveTurn.ts:63) is the **sole** agent-spawn site (`cline.ts`
+     only reads the static tool catalog).
+   - Binary forensics: the `claude` binary contains **26 osascript calls** incl.
+     `tell application "Terminal"/"iTerm"/"Electron"/"Finder"/"System Events"`.
+     Categorized all 26 — every real call is **user-action-gated** (clipboard
+     image, screenshot, `Apple_Terminal`-only theme read, notifications,
+     terminal-setup, Claude-Desktop deep link via `tell application "Electron"`,
+     browser-default-handler detection). The only path that fires **without user
+     action** when `claude` is launched inside the VS Code extension host is
+     **IDE auto-connect** (osascripts to identify/activate the editor). We never
+     use the IDE connection → declining is harmless = exact symptom.
+   - Gate confirmed: `if(!((autoConnectIde||…||CLAUDE_CODE_SSE_PORT||…||
+     CH(CLAUDE_CODE_AUTO_CONNECT_IDE)) && !mK(CLAUDE_CODE_AUTO_CONNECT_IDE))) return;`
+     with `mK("0")===true` → `CLAUDE_CODE_AUTO_CONNECT_IDE=0` hard-disables it.
+
+4. **Fix design (verified end-to-end, statically).** Pass `options.env` to the
+   Agent. The provider merges `{...getBaseProcessEnv(), ...settings.env}` and the
+   agent-SDK spawns the child with `env = L4 ? {...L4} : {...process.env}`. Seeding
+   `options.env` with the full `process.env` keeps default inheritance intact
+   (login under HOME, proxy/CA vars) and adds only `CLAUDE_CODE_AUTO_CONNECT_IDE=0`
+   (+ `CLAUDE_CODE_IDE_SKIP_AUTO_INSTALL=1`, belt-and-suspenders for the
+   extension-install deep link). `options` type is `Record<string,unknown>` — no
+   cast.
+
+5. **TDD.** RED: unit test asserts `runEditTurn` constructs the Agent with both
+   switches set and the rest of `process.env` preserved (fake Agent captures
+   `cfg`). Watched it fail (`env` undefined). GREEN: added `options.env` to the
+   Agent constructor in `liveTurn.ts`. 250 unit + typecheck + build green.
+
+6. **Empirical live-turn smoke (`scripts/smoke-live-turn.mjs` + osascript shim)
+   was DECLINED** by the operator (it makes a live network/auth'd agent call).
+   Env-preservation rests on the static chain proof + the unit test instead.
+
+7. **Ship.** Branch `s60-59-apple-events-prompt` → commit → push → **PR #68** →
+   squash-merge to main (`e671c4b`), branch deleted, **#59 auto-closed**.
+   Post-merge: 250 unit green on fresh main.
+
+## Cut state
+
+- **#59 SHIPPED** to main (squash `e671c4b`, PR #68; issue closed).
+- `src/liveTurn.ts` (+19) and `test/liveTurn.test.ts` (+38) only.
+- 250 unit + typecheck + build green; E2E unaffected (it stubs the turn, never
+  reaching the real spawn path).
+- No implementation-plan artifact (small leaf bug) → nothing to archive in
+  `plans/`. No deploy pipeline stage (VS Code extension, no flotilla/PPE) →
+  change ships at merge.
+
+## Deferred decisions
+
+- **Did not run the real-turn empirical smoke** (operator declined the live
+  agent call). Alternative: run it with an osascript shim to observe the trigger
+  directly. Why proceeded: the env→spawn chain and gate semantics are proven by
+  static analysis, the plumbing is unit-tested, and the definitive confirmation
+  (fresh-TCC GUI check) is inherently an operator gesture per the issue's
+  acceptance — so the smoke was not load-bearing for shipping.
+- **Set both `AUTO_CONNECT_IDE=0` and `IDE_SKIP_AUTO_INSTALL=1`** rather than the
+  single minimal switch. Alternative: auto-connect=0 alone (likely sufficient,
+  since auto-install rides the auto-connect flow). Why both: cheap, explicit,
+  both are no-ops outside an IDE; defense-in-depth against a separately-gated
+  install path.
+
+## Operator plate (next session)
+
+1. **Operator F5 smoke of #64** (F12 inline editor diff) — still pending from 0059.
+2. **Fresh-TCC GUI smoke of #59** — reset macOS Automation TCC for the host app,
+   exercise a Claude edit, confirm the "control other applications" prompt is
+   gone and the edit still works.
+3. Then a backlog item — **#58** (P2 story, self-contained) is the most actionable.
+
+## Next-session prompt
+
+```
+/goal Operator-smoke #64 (F5) + fresh-TCC GUI smoke of #59, then plan-and-execute #58 (proposal accept/decline word-level diff — route proposalBlockHtml through wordMergedMarkdown)
+```
@@ -1,13 +1,13 @@
-# Session 0021.0 — Transcript
+# Session 0061.0 — Transcript

 > App: vscode-cowriting-plugin
-> Start: 2026-06-11T08-33 (PST)
+> Start: 2026-06-27T05-14 (PST)
 > Type: planning-and-executing
 > Status: **PLACEHOLDER — claimed at session start; finalized at session end.**
 >
-> This file reserves session ID 0021 for vscode-cowriting-plugin. The driver replaces this
+> This file reserves session ID 0061 for vscode-cowriting-plugin. The driver replaces this
 > body with the full transcript and renames the file to its final
-> SESSION-0021.0-TRANSCRIPT-2026-06-11T08-33--<end>.md form at session end.
+> SESSION-0061.0-TRANSCRIPT-2026-06-27T05-14--<end>.md form at session end.

 ## Launch prompt

@@ -0,0 +1,23 @@
+# Session 0062.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-06-27T17-02 (PST)
+> Type: brainstorming
+> Posture: careful
+> Claude-Session: 2c832f9a-a70c-4d71-ab13-c4f39d6eb595
+> Checkout: /Users/benstull/git/benstull.org/benstull/vscode-cowriting-plugin
+> Status: **PLACEHOLDER — claimed at session start; finalized at session end.**
+>
+> This file reserves session ID 0062 for vscode-cowriting-plugin. The driver replaces this
+> body with the full transcript and renames the file to its final
+> SESSION-0062.0-TRANSCRIPT-2026-06-27T17-02--<end>.md form at session end.
+
+## Launch prompt
+
+_(launch prompt not captured at claim time)_
+
+## Deferred decisions
+
+_Autonomous-mode low-confidence calls the driver made and would have
+liked operator input on. Appended as the session runs; surfaced at
+finalize. Empty if none._
@@ -0,0 +1,65 @@
+# Session 0063.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-07-01T18-49 (PST)
+> Type: brainstorming
+> Posture: careful
+> Claude-Session: 1a8fc96b-6f36-4102-be7b-decb53251b2b
+> Checkout: /Users/benstull/git/benstull.org/benstull/vscode-cowriting-plugin-s0063
+> Status: **PLACEHOLDER — claimed at session start; finalized at session end.**
+>
+> This file reserves session ID 0063 for vscode-cowriting-plugin. The driver replaces this
+> body with the full transcript and renames the file to its final
+> SESSION-0063.0-TRANSCRIPT-2026-07-01T18-49--<end>.md form at session end.
+
+## Launch prompt
+
+```
+Examine the existing vs plugin, the prototype, and the half-baked spec that I used Opus to create and make a recommendation for a VS Code experience that allows humans and machines to collaborate on a document, with the machine making suggestions based on human input and the human able to approve, reject, or modify those suggestions. This should use as many VS Code UX paradigms as possible, use as many existing VS Code components or extensions, but create custom components/UX when needed. You can assume this is just Claude as the 'machine' to start
+
+```
+
+## Plan
+
+> Anchor: design-only north-star recommendation (no tracker anchor; lineage =
+> Epic #1 / F1–F12 + session 0062's `coauthoring-native-surfaces.md` draft —
+> same anchor treatment 0062 recorded)
+
+1. Examine the three inputs: the shipped plugin (F1–F12 code on `main`), the
+   prototype, and the half-baked Opus-authored spec (untracked `specs/` in the
+   primary checkout).
+2. Identify/confirm the anchoring issue (§4.3 R1 gate) — capture one if none fits.
+3. Walk the consequential UX decisions with the operator (AskUserQuestion
+   checkpoints), grounded in native VS Code paradigms vs custom components.
+4. Draft the recommendation as a Solution-Design spec in the content repo's
+   `specs/`; whole-document review at the Gitea render URL.
+
+## Session record (running)
+
+- Examined the three inputs: shipped plugin (F1–F12 inventory), prototype
+  (`vscode-cowriting-prototype` = rung-2 spike: E1/E3/E5 PASS — key finding:
+  CodeLens does not render in a diff editor; E2/E4 unbuilt), and the 0062 draft
+  `coauthoring-native-surfaces.md` v0.1.0 + companion spike plan.
+- Recommendation delivered + four forks decided with the operator
+  (AskUserQuestion checkpoints, all recommended options chosen):
+  1. Review model = **in-buffer pending changes + per-hunk CodeLens** (D18;
+     supersedes multi-diff Keep/Undo; amends INV-5/INV-12; resolves Q2/Q7).
+  2. Packaging (Q1) = **evolve the shipped extension in place** (D17; §6.10
+     rewritten from "greenfield").
+  3. Ask surface = **comments-first** (D19; split-pane input webview sunsets).
+  4. Artifact = **amend the 0062 spec to v0.2.0** on content-repo branch
+     `session-0063` (0062's working tree untouched). Also D20 (tweaks × per-hunk
+     decisions), D21 (D11 → shipped reality).
+- Spec v0.2.0 + spike-spec status committed (`431ebeb`) and pushed to
+  `session-0063`; whole-document review handed to the operator at the Gitea
+  render URL (terminal gate).
+
+## Deferred decisions
+
+- Proceeded past the concurrent-session warning (0061/0062 both INPROGRESS,
+  placeholder-only claims from 2026-06-27) because the operator explicitly
+  launched this session; isolated in worktree `session-0063` per #134/§5.4.
+
+_Autonomous-mode low-confidence calls the driver made and would have
+liked operator input on. Appended as the session runs; surfaced at
+finalize. Empty if none._
@@ -0,0 +1,58 @@
+# Session 0064.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-07-01T22-19 (PST) · End: 2026-07-02T05-34 (PST)
+> Type: writing-plans
+> Posture: careful
+> Claude-Session: 7c03a44e-1f92-4445-afd9-928efd55afba
+> Checkout: /Users/benstull/git/benstull.org/benstull/vscode-cowriting-plugin-wp (isolated worktree; torn down at finalize)
+> Anchor: design specs/coauthoring-native-surfaces.md (v0.2.1, graduated)
+
+## Launch prompt
+
+> Got it. Let's implement the full plugin
+> (operator, after driving the rung-2 feasibility spike; context: evolve the shipped
+> extension in place onto specs/coauthoring-native-surfaces.md, D17/§6.10/§7.1)
+
+## Pre-state
+
+- Native-surfaces Solution Design at **v0.2.0, draft**, on content-repo branch `session-0063` (session 0063 INPROGRESS, awaiting operator review). Content `main` had never received the 0062/0063 spec lineage.
+- Rung-2 feasibility spike: E1/E3/E5 PASS (recorded in v0.2.0); **E2/E4 unvalidated**. Earlier in this same conversation (pre-session), the throwaway prototype `vscode-cowriting-prototype` was rebuilt to drive the FULL v0.2.0 loop (enter → ask-via-comment → pending-in-buffer → tweak → baseline/diff → preview → stop) and the operator drove it end to end.
+- Three concurrent INPROGRESS sessions (0061/0062/0063); 0062 occupied the main plugin checkout → this session claimed from an isolated worktree (#134 gate).
+
+## Arc (turn by turn, condensed)
+
+1. **Spike driving (pre-claim, same conversation):** operator drove the walkthrough; two E4 defects surfaced and were fixed live in the prototype — (a) "Make this edit" gated on a nonexistent `commentThreadContextValue` key (real key: **`commentThread`**); (b) Ask-Claude didn't focus the comment input (fixed via **`workbench.action.addComment`**); (c) "cursor must be within a commenting range" → commenting ranges are **cached per document**; re-assigning `commentingRangeProvider` is the only refresh signal. All recorded in the prototype's FINDINGS.md.
+2. **Gate + claim:** operator asked to "implement the full plugin" → `wgl-writing-plans` init; claim refused the shared checkout (live 0062) → isolated worktree `vscode-cowriting-plugin-wp`, claimed **0064** (`--type writing-plans`, careful).
+3. **Eligibility:** `check-issue-eligibility.sh --for implementation-plan --design …coauthoring-native-surfaces.md` → **ELIGIBLE (R2a)**. Design-readiness: N/A (spec §5.1 = Code directly, no UI design).
+4. **Sequencing decision (operator):** fold spike findings first, then plan. Operator confirmed the drive reached step 5+ → **E2 PASS**; E4 **PASS-with-fallback**.
+5. **Spec v0.2.1 delta** (content worktree branch `session-0064` off `session-0063`): §6.4 spike-verified API seams (commenting-range refresh, addComment focus, `commentThread` when-key, offer affordances; baseline nested-repo + `state.HEAD.commit` notes), §7.1 rung 2 closed (E2 PASS / E4 PASS-with-fallback; D15 preview-selection half explicitly un-spiked), §7.3 risks updated. Pushed; operator **approved → graduated**.
+6. **Plan written** (`superpowers:writing-plans`; codebase mapped via an Explore subagent + direct reads): `docs/superpowers/plans/2026-07-01-native-surfaces-migration.md` — ONE plan, 9 tasks in §7.1 rung-4 order: (1) CoeditingRegistry (INV-10) → (2) baseline router git-HEAD+snapshot (INV-7; **retires INV-18** machine-landing advance) → (3) QuickDiff + `cowriting-baseline:` + Review Changes + status bar (INV-13) → (4) gate every surface → (5) extract EditFlow from the review webview → (6) comments-first ask + comment→reply→offer→proposal loop (D19/D10/D8; sunsets the input webview) → (7) built-in-preview annotations via markdown-it sentinels (D3/D21; Q4 mermaid stop-and-record check) → (8) sunset the review webview + ✓ Keep/✗ Reject copy + README → (9) full-loop E2E + real-SDK smoke (rung 3 folded in). D15 explicitly deferred. Committed to plugin branch `session-0064`, pushed.
+7. **Review gate:** operator approved — explicitly accepting the **D15 deferral** and the **INV-18 retirement** — and directed execution to a new session.
+8. **Finalize (single approved plan):** content `session-0064` merged → content `main` (landed the whole 0062/0063/0064 spec lineage; flagged), plugin `session-0064` fast-forwarded → plugin `main`; plan archived via `submit-plan.sh` → content `plans/2026-07-01-native-surfaces-migration.md` (ddaf3b0); **issue #69** filed (D15, feature P2, resume context in body); both session worktrees removed; memory updated.
+
+## Deferred decisions
+
+_None — the two judgment calls (D15 deferral, INV-18 retirement) were put to the operator explicitly and approved at the plan gate._
+
+## Cut state
+
+- **Spec:** `coauthoring-native-surfaces.md` **v0.2.1 `graduated`**, on content `main`.
+- **Plan:** approved + archived at content `plans/2026-07-01-native-surfaces-migration.md`; also on plugin `main` under `docs/superpowers/plans/`. **Not executed** (by design — writing-plans is terminal).
+- **Spike:** complete (5/5); prototype **retained deliberately** (operator: iterate UX there before the real build) — supersedes the spike-plan's delete-when-done for now.
+- **Tracker:** #69 (D15 preview-initiated comments, feature P2) parked with full resume context.
+- Leftover: merged-but-undeleted local branch refs (`session-wp-native-surfaces` in the plugin clone, `session-0064` in the content clone) — operator declined force-deletion; plain `-d` from a main checkout clears them. Sessions 0061/0062/0063 remain INPROGRESS (0062/0063 will find their content branches already merged).
+
+## Operator plate
+
+- Open the executing session (below) when ready.
+- Sessions 0061/0062/0063 are still unfinalized — worth closing out.
+- Optional tidy: delete the two stale local branch refs.
+
+## Next session
+
+```
+/goal Execute the native-surfaces migration plan (content repo plans/2026-07-01-native-surfaces-migration.md, 9 tasks, from graduated spec coauthoring-native-surfaces.md v0.2.1) end to end
+```
+
+(Open with `wgl-executing-plans`; the plan's tasks run 1→9 strictly, each ending typecheck+unit+host-E2E green; Task 9's live smoke is the operator-run rung-3 gate.)
@@ -0,0 +1,68 @@
+# Session 0065.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-07-02T06-31 (PST)
+> End: 2026-07-02T16-10 (PST)
+> Type: executing-plans
+> Posture: yolo
+> Claude-Session: 32dcda47-7361-4fa7-8145-9d118b5de3cc
+> Checkout: /Users/benstull/git/benstull.org/benstull/vscode-cowriting-plugin/.claude/worktrees/session-native-surfaces-exec (isolated worktree, branch `session-0065`)
+> Status: FINALIZED
+
+## Launch prompt
+
+> The last session should have set a goal. If not, find the spec that we should be implementing
+
+## Plan
+
+- **Anchor:** graduated Solution Design `coauthoring-native-surfaces.md` v0.2.1 (content repo, on main) — R2(a), ELIGIBLE per `check-issue-eligibility.sh`; lineage: sessions 0062 (spec) → 0063 (v0.2.0) → 0064 (spike + plan approved).
+- **Goal:** execute the reviewed implementation plan `plans/2026-07-01-native-surfaces-migration.md` — 9 tasks in §7.1 rung-4 order — and land it (no §9 web pipeline: VS Code extension, spec §7.2; done = merged + suites green + rung-3 smoke).
+- **Isolation:** sessions 0061/0062/0063 live-concurrent; 0062 occupied the primary checkout → claim refused shared tree (plugin #134) → EnterWorktree `session-native-surfaces-exec`, branch renamed `session-0065`.
+- Approved deliberate changes carried by the plan: INV-18 retirement (no machine-landing baseline advance); D15 deferred as issue #69.
+
+## Pre-state
+
+- Plugin repo `main` at `ef61b14` (0064's plan-archive commit); clean; 265 unit + full host E2E green at baseline (verified before any work).
+- Content repo `main` at `ddaf3b0` (0064 lineage merged); local content clone behind origin + carrying other sessions' uncommitted drafts — never touched; spec read via `git show origin/main:`.
+
+## Arc (turn-by-turn, compressed)
+
+1. **Init:** goal recovered from memory (0064 note). Claim dry-run → shared-checkout refusal → worktree isolation → claimed **0065** (`26b2212`). Anchor gate: ELIGIBLE (R2a). Baseline verified green.
+2. **Execution model:** superpowers:subagent-driven-development — fresh implementer subagent per task, task-scoped reviewer (spec + quality verdicts), fix subagents on findings, re-review loops; durable ledger `.superpowers/sdd/progress.md`; final whole-branch review at the end. Models tiered (sonnet implementers; opus for risky-diff reviews; fable for the final review).
+3. **Task 1 — CoeditingRegistry (INV-10 gate):** `2597cba`; review found untested `syncContext` context-key derivation (plan-mandated gap) → fix `de83757` (11 unit tests). Approved.
+4. **Task 2 — baseline router (INV-7):** `447a117`. Deviations approved by review: `.git/logs/HEAD` reflog watcher nudging `repo.status()` (vscode.git's watcher missed external commits in the host; change detection still keyed off `state.HEAD.commit`), E2E suites gaining enter-calls early. Machine-landing advance deleted; `markReviewed` replaces `pinDiffBaseline`. Approved.
+5. **Task 3 — native diff surface:** `cf65528` (QuickDiff, `cowriting-baseline:` provider, Review Changes, status bar, pure `countLineHunks`). Approved.
+6. **Task 4 — gate every surface:** `a323b82`; review confirmed a PUC-7 restore-on-enter gap (proposals/attribution) → fix `a36353d` (registry subscribers + regression E2E verified fail-without-fix). Approved.
+7. **Plan-file repair:** fused `---### Task 5` heading broke the task extractor → `f23fc4a`.
+8. **Task 5 — EditFlow extraction:** `ad5d34b`; review caught a real regression (palette accept/reject-all no longer repainted an open review panel) → fix `8fbbe45`. Approved.
+9. **Task 6 — comments-first ask loop:** `75f4a3c`; review found a **Critical**: `cowriting.edit`'s no-selection route (⌘⌥E) dead-ended in the rejecting input stub → fix `c9975ba` (all ask entry points converge on `ThreadController.askClaude`; machine-author guard; offer-WeakMap cleanup). Approved. `editInstructionInput.ts` deleted.
+10. **Task 7 — built-in preview annotations:** `17fc01e` (pure markdown-it plugin; sentinel machinery generalized 3-way, backward-compatible — 84/84 old tests unedited; render-wrapper approach correctly replacing the brief's per-token rule; mermaid rendering via `markdown.previewScripts` — Q4 confirmed). Approved; intra-diagram diff scoped out by the implementer.
+11. **Task 8 — webview sunset:** `2170a0d` (−1472 lines; coverage moved in-commit, arithmetic reconciled exactly). Review flagged the Task-7 scope-out as an unsanctioned INV-29 retirement + `renderReview` going production-dead. **Adjudication:** plan T7 §2.6 mandated the re-emit and its stop condition never triggered → fix `705de31` (diffMermaid wired into the fence path, 6 ported unit tests) + hardening `ff42b5a` (body-matched queue; misalignment degrades to verbatim; parse-less re-render reset). `renderReview` kept deliberately (see Deferred decisions). Approved.
+12. **Task 9 — full-loop E2E + smoke:** `22c9da0` (PUC-7→8→2→1 walk; `scripts/smoke-native-loop.mjs` + `npm run smoke:native`, operator-run). Review verified the implementer's adjudication (interior tweak → anchor orphans by contract; Keep-by-id survives) and surfaced a **pre-existing INV-5 gap** (reject-after-interior-edit) → plan text corrected (`17366a4`) + **issue #70 filed** (bug P2).
+13. **Final whole-branch review (fable):** verdict "with fixes" — Critical (preview-annotation host leaked past INV-10: wrong-doc annotation + stale-after-exit), Important (thread contextValue conflation made Resolve/Reopen permanently unmatchable once Claude replied; coedited docs opened post-activation never got a baseline) → ONE fix wave `0d69a29` (all three at root + regression E2Es + fold-in minors + polish). Re-review: full when-clause state matrix verified, no double-establish race, **Ready to merge: Yes**. **Issue #71 filed** (surface-polish batch).
+14. **Ship:** final verification (typecheck, 301 unit, 90+5 E2E, build — all green) → pushed → **PR #72** → **squash-merged to main `935fcc3`** → remote branch deleted → plan re-archived to content repo (`b0687b6`).
+
+## Deferred decisions
+
+- **`renderReview` left in place though production-dead after the Task 8 webview sunset.** The old F10 pure render engine (`renderReview`/`renderOp` in `trackChangesModel.ts`) lost its last production consumer when `trackChangesPreview.ts` was deleted; three E2E suites (authorship, s48PinClean, undoMarks) now probe it directly as a pure module via a test helper. I kept it (tested pure module, INV-33 documentation value, low carrying cost) rather than deleting + reworking those suites mid-migration. Operator may prefer deletion as a follow-up cleanup.
+- **INV-29 mermaid-diff parity adjudicated as plan-mandated, not retired.** Task 7 scoped out the F7.1 re-emit for testability; the plan's stop condition (previewScripts unworkable) never triggered, so I dispatched a fix to wire `mermaidDiff` into the preview fence path rather than accepting the silent retirement. Judged from plan T7 §2.6 + the plan's spec-coverage self-check.
+
+## Cut state
+
+- Plugin `main` = `935fcc3` (PR #72 squash). Working tree clean. Remote `session-0065` deleted; **local branch ref `session-0065` remains** (squash → plain `-d` refuses; operator has preferred manual cleanup of such refs).
+- Content repo `main` = `b0687b6` (plan re-archived with in-execution edits: heading fix + Task 9 wording correction).
+- Suites at HEAD: typecheck · 301 unit · 90 + 5 host E2E (1 env-conditional #54 undo-probe pending on some runs) · esbuild — all green.
+- Issues: **#70 filed** (INV-5 reject-after-interior-edit, bug P2), **#71 filed** (surface polish, task P3); #69 (D15) pre-existing.
+- Sessions 0061/0062/0063 still INPROGRESS (other Claude sessions); their checkouts/branches untouched.
+
+## Operator plate
+
+1. **`npm run smoke:native`** — rung-3 real-SDK smoke through the comment loop (one green run completes the plan's definition of done; report it to this transcript's home if you want the record complete).
+2. **Mermaid-preview GUI smoke** — open a coedited doc with a changed ```mermaid fence in the built-in preview; confirm diagram renders with cwAdded/ghost styling + legend (previewScripts path isn't E2E-queryable).
+3. Optional cleanup: `git branch -D session-0065` in the plugin clone; delete the now-merged worktree leftovers if any; #71 when convenient.
+
+## Next-session prompt
+
+```
+/goal Plan and execute #70 — fix the INV-5 reject-after-interior-edit gap (proposalController revertInPlace orphan path); fix directions in the issue body
+```
@@ -0,0 +1,113 @@
+# Session 0066.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-07-02T16-24 (PST)
+> End: 2026-07-02T18-54 (PST)
+> Type: planning-and-executing
+> Posture: yolo
+> Claude-Session: 14c9f9cc-92a1-47ef-a12d-9b0ac2469e13
+> Checkout: /Users/benstull/git/benstull.org/benstull/vscode-cowriting-plugin/.claude/worktrees/fix-70-inv5-reject-gap
+> Status: FINALIZED
+
+## Launch prompt
+
+```
+/goal Plan and execute #70 — fix the INV-5 reject-after-interior-edit gap (proposalController revertInPlace orphan path); fix directions in the issue body
+```
+
+## Plan
+
+> Anchor: #70 (type/bug, P2 — ELIGIBLE R2b)
+
+Fix the INV-5 reject-after-interior-edit gap: `revertInPlace` in
+`src/proposalController.ts` skipped the restore when an interior edit inside
+the pending proposal's range orphaned the exact-substring anchor (INV-11),
+yet removed the proposal and reported success — ✗ Reject silently left the
+(edited) proposed text in the buffer. Evaluate the issue's fix directions
+(a/b/c), TDD the fix, ship branch → PR → merge per §5.4.
+
+## Pre-state
+
+- `origin/main` at `d8f2d73` (session 0065's native-surfaces migration
+  shipped; #70/#71 filed from its final review).
+- Three concurrent sessions in flight (0061, 0062, 0063 placeholders);
+  session 0062 occupied the canonical checkout, so the #134 claim guard
+  refused a shared-tree claim — this session created worktree
+  `.claude/worktrees/fix-70-inv5-reject-gap` off `origin/main` FIRST and
+  claimed 0066 from inside it (the inverted order the init documents).
+- Canonical checkout's untracked files (`specs/`, a plan doc) belong to the
+  concurrent session and were left untouched.
+
+## Arc
+
+1. **Init.** Claimed 0066 from the worktree; #70 checked ELIGIBLE (type/bug,
+   R2b; needed `--host git.benstull.org` + the issues token service). Fetched
+   the issue body carrying the three fix directions.
+2. **Design.** Explored `proposalController.ts` / `anchorer.ts` /
+   `editorProposalController.ts`. Chose direction **(a)** — tracked applied
+   range — with **(b)** honest hard failure as fallback; rejected **(c)**
+   fuzzy resolve (INV-11: never guess). Key insight: `state.live` is
+   shift-maintained but clobbered to a collapsed placeholder by any
+   intervening `renderAll`, so a dedicated `appliedSpans` map with explicit
+   trust semantics was designed instead.
+3. **Plan.** `docs/superpowers/plans/2026-07-02-70-inv5-reject-after-interior-edit.md`
+   (4 TDD tasks). Fused single-session path (leaf bug), executed inline.
+4. **Tasks 1–4** (commits `a9f8dd0`, `82cea74`, `ddd634b`, `f47dfbc`):
+   appliedSpans + revertInPlace fallback (E2E repro failed-then-passed);
+   hard-failure lock test (first cut used a whole-buffer rewrite and FAILED —
+   VS Code minimizes workspace edits, so it decomposed into interior hunks
+   the span legitimately survives; switched to a boundary-straddling
+   deletion); rejectAll `{reverted, skipped}` parity; fullLoop reject leg now
+   tweaks inside the pending range first (retiring its deliberate avoidance).
+5. **Review wave** (high-effort multi-agent workflow; 10 CONFIRMED findings,
+   9 fixed in `e389406`): pure `shiftTracked` in anchorer (+11 unit tests;
+   insertion-at-span-start lands BEFORE the span), tracked spans cleared on
+   doc close (disk can change with no events), rebuild-only resync (duplicate
+   -text resolve can't clobber continuous tracking), `guard.isReadOnly` on
+   finalize/revert (INV-16), "Discard proposal (leave text)" action on the
+   hard-fail warning (reload leaves no other dismissal path), CodeLens pair
+   anchored at the tracked span (the fixed path stays reachable from the
+   primary gesture), QuickPick batch menu routed through the reporting
+   commands, one `clearProposal` helper (fixes `reject()`'s stale bookkeeping).
+   The 10th finding is pre-existing → filed **#74**.
+6. **Ship.** PR #73 → squash-merged to `main` as `7583165`; #70 auto-closed.
+   312 unit + 94/5 host E2E green (1 occasionally pending = the known #54
+   undo-capability probe). Plan archived to the content repo
+   (`plans/2026-07-02-70-inv5-reject-after-interior-edit.md` at `0019d58`).
+   Remote branch `s70-inv5-reject-gap` deleted after merge.
+
+## Deployment pipeline (§9)
+
+This app is a VS Code extension with no deploy infra (no flotilla/PPE stage
+exists — per prior sessions, "no deploy pipeline (extension, no flotilla)").
+The localhost stage ran in full: typecheck + 312 unit + 94/5 host E2E green.
+Nothing to deploy; no release tag applies.
+
+## Cut state
+
+- `origin/main` = `7583165` (the #70 fix). Worktree branch merged; remote
+  branch deleted; worktree torn down at finalize.
+- Issue #70 CLOSED (auto, via PR). Issue #74 (bug P3) OPEN — filed this
+  session: `resolve()` accepts a single exact occurrence without a context
+  check (pre-existing; duplicated block can hijack any anchor consumer).
+- Concurrent sessions 0061/0062/0063 still INPROGRESS — not mine to adopt
+  (live, per their stamps).
+
+## Operator plate
+
+- **Manual smoke owed:** the "Discard proposal (leave text)" warning button
+  (native notification — not E2E-testable) + the smokes already owed from
+  sessions 0062/0065.
+- Prod promotion: N/A (no deploy pipeline for the extension).
+
+## Deferred decisions
+
+No low-confidence calls this session. (Design forks — (a)+(b) over (c), the
+distrust semantics, squash-merge — all followed the issue's own directions,
+existing invariants, or repo convention.)
+
+## Next-session prompt
+
+```
+/goal Plan and execute #71 — surface-polish batch from the native-surfaces final review (6 small items listed in the issue body)
+```
@@ -0,0 +1,109 @@
+# Session 0067.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-07-02T20-07 (PST)
+> Type: planning-and-executing
+> Posture: yolo
+> Claude-Session: d29a9976-cb2d-427b-9ef1-4488b3c2a262
+> Checkout: /Users/benstull/git/benstull.org/benstull/vscode-cowriting-plugin/.claude/worktrees/s71-surface-polish
+> End: 2026-07-03T07-39 (PST)
+> Status: **FINALIZED**
+
+## Launch prompt
+
+```
+Plan and execute #71 — surface-polish batch from the native-surfaces final review (6 small items listed in the issue body)
+
+```
+
+## Plan
+
+> Anchor: #71 (type/task, P3 — ELIGIBLE R2b)
+
+Plan and execute #71 — the surface-polish batch from the native-surfaces
+migration final review (session 0065). Six small items from the issue body:
+
+1. `markReviewed`/`pin()` misleading warning on never-established docs (`src/diffViewController.ts`)
+2. Delete orphaned `cowriting.acceptProposal`/`rejectProposal` declarations (`package.json`)
+3. `openReviewPreview` editor/title when-key: `editorLangId` → `resourceLangId`
+4. README F3 section drift: SUPERSEDED banner + stale chord references
+5. Gate/warn `cowriting.createThread` palette entry when not coediting
+6. Clear `gitBaseline` reflog debounce timer in `dispose()` (`src/gitBaseline.ts`)
+
+Fused session (trivial leaf batch): just-in-time plan → execute → tests green
+→ PR → merge. Isolated worktree `s71-surface-polish` (checkout contended by
+live session 0062).
+
+## Execution record
+
+- Plan written (`docs/superpowers/plans/2026-07-02-71-surface-polish.md`, committed
+  on the branch) and executed inline — all six items landed as one commit each:
+  1. `pin()` not-coediting branch → "Run ✦ Coedit this Document with Claude first."
+  2. orphaned `acceptProposal`/`rejectProposal` declarations + `when:false` palette
+     entries deleted
+  3. `openReviewPreview` editor/title when-key `editorLangId` → `resourceLangId`
+     (+ the E2E assertion at `reviewPanelMenu.test.ts:129` that codified the old
+     key — the one mid-run E2E failure, fixed forward)
+  4. README: F3 superseded-commands banner; F7/F10 banner chord notes
+  5. `createThread` palette entry gated (same when as its editor/context entry)
+  6. `gitBaseline` reflog debounce timer cleared on dispose
+- Verification: 312 unit + typecheck + build + full host E2E green.
+- Shipped: PR #75 squash-merged to `main` as `3e7ad58`; issue #71 auto-closed;
+  remote branch deleted. No flotilla/PPE pipeline — this app is a VS Code
+  extension with no deployed service (§9 ends at merge + green suites here).
+
+## Post-ship arc (operator review + Q&A)
+
+- Launched the EDH from the worktree for operator review
+  (`code --extensionDevelopmentPath=<worktree> <worktree>/sandbox`; VS Code
+  found at `~/Downloads/Visual Studio Code.app`).
+- Operator Q&A, all assessment-only: (1) comment box is the native Comments
+  API widget — Cmd+Enter submits, Enter newlines; operator chose to keep the
+  native mapping. (2) macOS "access data from other apps" TCC prompt — App
+  Data Protection category, distinct from #59; VS Code found QUARANTINED +
+  App-Translocated (runs from ~/Downloads) — recommended move to
+  /Applications; #59 env guard confirmed present (`liveTurn.ts:82-83`).
+  (3) External (outside-editor) Claude edits: detected via the disk-sync
+  branch (`attributionController.ts:175`), rendered neutral, never attributed
+  — cross-rung sidecar is the designed path to attributed external edits.
+  (4) `onDidChangeTextDocument` carries no originator identity (why the seam
+  pre-registers edits, INV-9). (5) Sketched per-author diff surfaces
+  (synthetic per-author baselines + second QuickDiffProvider). (6) Explained
+  tweak-before-accept tracking (mixed authorship; #70 appliedSpans; reject
+  reverts whole span or refuses).
+- The Q&A converged on a product pivot the operator dictated: simplify review
+  to baseline-diff Keep/Reject (three-verb toolbar, no annotations, no
+  preview). Handed off to capture session 0068 (draft written, NOT filed) and
+  an implementation prompt.
+
+## Deferred decisions
+
+- design-readiness soft gate returned MISSING for #71 — proceeded: all six
+  items are copy/menu-gating/cleanup within existing surfaces, no new UI
+  beyond the design system (R4 not implicated).
+
+## Closeout
+
+Shipped: PR #75 (squash `3e7ad58`), issue #71 closed, remote branch deleted.
+Suites: 312 unit + typecheck + build + full host E2E green (one mid-run E2E
+fix: `reviewPanelMenu.test.ts:129` had codified the wrong when-key). Plan
+archived to content repo `plans/2026-07-02-71-surface-polish.md`. No
+flotilla/PPE stage exists for this app (VS Code extension, no deployed
+service) — §9 ends at merge + green suites. Worktree `s71-surface-polish` is
+merged-clean but KEPT at finalize: the operator's EDH review window still runs
+from it (`--extensionDevelopmentPath`); remove after closing that window
+(`git worktree remove .claude/worktrees/s71-surface-polish`). Operator plate:
+close the EDH then remove the worktree; move VS Code.app out of ~/Downloads to
+/Applications (translocation/TCC); decide on the 0068 unfiled draft.
+
+## Next-session prompt
+
+/goal Resume the capture: file the unfiled draft
+issues/simplify-review-baseline-diff-keep-reject.md as type/feature P1 on
+benstull/vscode-cowriting-plugin, then brainstorm its Solution Design
+(specs/coauthoring-baseline-diff-review.md) on the new issue, locking the five
+forks in the draft's Solution notes.
+
+_Autonomous-mode low-confidence calls the driver made and would have
+liked operator input on. Appended as the session runs; surfaced at
+finalize. Empty if none._
@@ -0,0 +1,50 @@
+# Session 0068.0 — Transcript
+
+> App: vscode-cowriting-plugin
+> Start: 2026-07-03T06-53 (PST)
+> Type: capture
+> Posture: careful
+> Claude-Session: d29a9976-cb2d-427b-9ef1-4488b3c2a262
+> Checkout: /Users/benstull/git/benstull.org/benstull/vscode-cowriting-plugin/.claude/worktrees/s71-surface-polish
+> End: 2026-07-03T07-39 (PST)
+> Status: **FINALIZED — tracked-lite (capture); NOTHING FILED**
+
+## Launch prompt
+
+```
+Capture one type/feature issue: simplify the review model to baseline-diff Keep/Reject — comments to ask Claude, native diff to review (per-hunk Keep/Reject/Modify, Keep advances the baseline), no annotations, no preview surface; toolbar reduces to Ask Claude / Review Changes / Mark All Changes as Reviewed. Context developed in conversation following session 0067; brainstorming session to follow on the filed issue.
+
+```
+
+## Arc
+
+- Claimed 0068 (capture, careful) from the s71 worktree; resolved content repo
+  `benstull/vscode-cowriting-plugin-content`. Detector listed ~24 pre-#80
+  unfiled-marker drafts for long-filed issues (operator-owned backlog hygiene
+  — untouched).
+- Drafted `issues/simplify-review-baseline-diff-keep-reject.md` per the
+  feature batching profile: business block → operator approved ("Looks
+  good") → structural block + solution notes (approach pre-validated in the
+  originating 0067 conversation; five design forks recorded).
+- **Filing withheld (INV-1):** at the file-on-approval gate the operator
+  pivoted — "Just give me the prompt for implementing" — so NO issue was
+  filed. Delivered the implementation prompt (unattended design→build chain
+  with named yolo override, plus the two-step interactive variant), with `#N`
+  pending the eventual filing.
+- Finalize: draft left in the content-repo working tree as the resumable
+  surface (never auto-discarded); type-fit judge not run (nothing filed);
+  type-fit: judged 0, mismatch 0, overridden 0.
+
+## Deferred decisions
+
+_None — careful posture; every gate was operator-answered. The one open
+operator decision (file the draft or not) is deliberately left open and
+carried in the next-session prompt._
+
+## Next-session prompt
+
+/goal Resume the capture: file the unfiled draft
+issues/simplify-review-baseline-diff-keep-reject.md as type/feature P1 on
+benstull/vscode-cowriting-plugin, then brainstorm its Solution Design
+(specs/coauthoring-baseline-diff-review.md) on the new issue, locking the five
+forks in the draft's Solution notes.
@@ -61,5 +61,146 @@
  },
  "0021": {
    "title": ""
+  },
+  "0022": {
+    "title": ""
+  },
+  "0023": {
+    "title": ""
+  },
+  "0024": {
+    "title": ""
+  },
+  "0025": {
+    "title": ""
+  },
+  "0026": {
+    "title": ""
+  },
+  "0027": {
+    "title": ""
+  },
+  "0028": {
+    "title": ""
+  },
+  "0029": {
+    "title": ""
+  },
+  "0030": {
+    "title": ""
+  },
+  "0031": {
+    "title": ""
+  },
+  "0032": {
+    "title": ""
+  },
+  "0033": {
+    "title": ""
+  },
+  "0034": {
+    "title": ""
+  },
+  "0035": {
+    "title": ""
+  },
+  "0036": {
+    "title": ""
+  },
+  "0037": {
+    "title": ""
+  },
+  "0038": {
+    "title": ""
+  },
+  "0039": {
+    "title": ""
+  },
+  "0040": {
+    "title": ""
+  },
+  "0041": {
+    "title": ""
+  },
+  "0042": {
+    "title": ""
+  },
+  "0043": {
+    "title": ""
+  },
+  "0044": {
+    "title": ""
+  },
+  "0045": {
+    "title": ""
+  },
+  "0046": {
+    "title": ""
+  },
+  "0047": {
+    "title": ""
+  },
+  "0048": {
+    "title": ""
+  },
+  "0049": {
+    "title": ""
+  },
+  "0050": {
+    "title": ""
+  },
+  "0051": {
+    "title": ""
+  },
+  "0052": {
+    "title": ""
+  },
+  "0053": {
+    "title": ""
+  },
+  "0054": {
+    "title": ""
+  },
+  "0055": {
+    "title": ""
+  },
+  "0056": {
+    "title": ""
+  },
+  "0057": {
+    "title": ""
+  },
+  "0058": {
+    "title": ""
+  },
+  "0059": {
+    "title": ""
+  },
+  "0060": {
+    "title": ""
+  },
+  "0061": {
+    "title": ""
+  },
+  "0062": {
+    "title": ""
+  },
+  "0063": {
+    "title": ""
+  },
+  "0064": {
+    "title": ""
+  },
+  "0065": {
+    "title": ""
+  },
+  "0066": {
+    "title": ""
+  },
+  "0067": {
+    "title": ""
+  },
+  "0068": {
+    "title": ""
  }
 }
@@ -101,6 +101,30 @@ export function resolve(docText: string, fp: Fingerprint): OffsetRange | "orphan
  return "orphaned";
 }

+/**
+ * #70 (INV-5/INV-11): maintain a TRACKED applied span across an in-session edit.
+ * Unlike `shift`, which clamps every point into a best-effort range, a tracked
+ * span must stay *provably* meaningful — it is a revert target. So:
+ *   - an edit fully OUTSIDE or fully INSIDE the span shifts it (the span is
+ *     still "the applied block, as tweaked");
+ *   - an insertion exactly AT the span start lands BEFORE the span (both ends
+ *     shift right) — matching exact-resolve semantics, where text typed just
+ *     before the applied block is never part of it;
+ *   - any edit STRADDLING a span boundary returns "distrusted": the shifted
+ *     result would be a clamped guess, and stale text is never reverted by
+ *     guess (INV-11).
+ */
+export function shiftTracked(range: OffsetRange, edit: TextEdit): OffsetRange | "distrusted" {
+  const delta = edit.newLength - (edit.end - edit.start);
+  if (edit.start === edit.end && edit.start === range.start) {
+    return { start: range.start + delta, end: range.end + delta };
+  }
+  const outside = edit.end <= range.start || edit.start >= range.end;
+  const inside = edit.start >= range.start && edit.end <= range.end;
+  if (!outside && !inside) return "distrusted";
+  return shift(range, edit);
+}
+
 /** Maintain a live range across an in-session edit (spec §6.4 `shift`). */
 export function shift(range: OffsetRange, edit: TextEdit): OffsetRange {
  const delta = edit.newLength - (edit.end - edit.start);
@@ -2,21 +2,26 @@
 * AttributionController — the thin editor-facing layer for F3 (spec §6.2).
 * Wires the pure AttributionTracker + PendingEditRegistry + Store/Anchorer to
 * the editor: human typing → human spans, seam edits → agent spans (INV-9),
- * decorations (Claude tint / human gutter border / toggle), save-time
- * persistence, load/external-change resolve-or-orphan (INV-1/INV-6), and the
- * orphan status-bar count. Sidecar self-write suppression and the shared
+ * save-time persistence, load/external-change resolve-or-orphan
+ * (INV-1/INV-6), and the orphan status-bar count. The editor carries no
+ * in-editor attribution decorations — the rendered preview is the single
+ * review surface (F10/INV-32); spansFor() feeds it the live attribution.
+ * Sidecar self-write suppression and the shared
 * FileSystemWatcher live in CoauthorStore / extension.ts (the sidecar is
 * co-owned with ThreadController).
 */
 import * as fs from "node:fs";
 import * as vscode from "vscode";
-import { CoauthorStore } from "./store";
+import { SidecarRouter, docIdentity } from "./sidecarRouter";
 import { newId, type AttributionRecord, type Provenance } from "./model";
 import { buildFingerprint, resolve, type OffsetRange } from "./anchorer";
 import { gitUserEmail } from "./identity";
 import { applyChange, coalesce, type LiveSpan } from "./attributionTracker";
 import { minimizeReplace, PendingEditRegistry } from "./pendingEdits";
 import type { VersionGuard } from "./versionGuard";
+import { isAuthorable } from "./workspacePath";
+import type { AuthorSpan } from "./trackChangesModel";
+import type { CoeditingRegistry } from "./coeditingRegistry";

 /** Test-facing snapshot of live attribution state for a document. */
 export interface RenderedSpan {
@@ -42,59 +47,71 @@ interface DocAttribution {
  hadAttributions: boolean;
 }

-const AGENT_DECO: vscode.DecorationRenderOptions = {
-  backgroundColor: "rgba(99, 102, 241, 0.18)",
-  overviewRulerColor: "rgba(99, 102, 241, 0.8)",
-  overviewRulerLane: vscode.OverviewRulerLane.Right,
-};
-const HUMAN_DECO: vscode.DecorationRenderOptions = {
-  borderColor: "rgba(16, 185, 129, 0.8)",
-  borderStyle: "solid",
-  borderWidth: "0 0 0 2px",
-};
-
 export class AttributionController implements vscode.Disposable {
  private readonly disposables: vscode.Disposable[] = [];
  private readonly docs = new Map<string, DocAttribution>();
  private readonly pending = new PendingEditRegistry();
-  private readonly agentType = vscode.window.createTextEditorDecorationType(AGENT_DECO);
-  private readonly humanType = vscode.window.createTextEditorDecorationType(HUMAN_DECO);
  private readonly statusItem = vscode.window.createStatusBarItem(vscode.StatusBarAlignment.Right, 90);
  private readonly output = vscode.window.createOutputChannel("Cowriting Attribution");
-  private visible = true;

  /**
   * F6 (§6.2/§6.4): the single machine-landing signal. Fired after a real
   * (non-no-op) seam apply succeeds. INV-9 makes the seam the sole machine-edit
-   * ingress, so this is the sole signal — DiffViewController subscribes to
-   * advance the baseline; no call-site wiring, no future driver can forget it.
+   * ingress, so this is the sole signal. Native-surfaces migration (INV-7/
+   * INV-18 retirement, spec §6.4): DiffViewController no longer subscribes
+   * here to advance the baseline — a landed edit stays a visible
+   * change-since-baseline until commit (head mode) or "Mark Changes as
+   * Reviewed" (snapshot mode). Other consumers may still subscribe.
   */
  private readonly applyEmitter = new vscode.EventEmitter<{ document: vscode.TextDocument }>();
  readonly onDidApplyAgentEdit: vscode.Event<{ document: vscode.TextDocument }> = this.applyEmitter.event;

  constructor(
-    private readonly store: CoauthorStore,
-    private readonly rootDir: string,
+    private readonly store: SidecarRouter,
+    private readonly rootDir: string | undefined,
    private readonly guard: VersionGuard,
+    private readonly registry: CoeditingRegistry,
  ) {
-    this.disposables.push(this.agentType, this.humanType, this.statusItem, this.output, this.applyEmitter);
+    this.disposables.push(this.statusItem, this.output, this.applyEmitter);
    this.disposables.push(
-      vscode.commands.registerCommand("cowriting.toggleAttribution", () => this.toggle()),
      vscode.workspace.onDidChangeTextDocument((e) => this.onDidChange(e)),
      vscode.workspace.onDidSaveTextDocument((d) => this.onDidSave(d)),
      vscode.window.onDidChangeActiveTextEditor(() => this.renderActive()),
+      // PUC-7 restore-on-enter: `loadAll` is the only path that (re)populates
+      // s.spans/s.orphans from the sidecar — without this, a document's FIRST
+      // entry into coediting (registry.enter(), extension.ts) never runs it,
+      // so pre-existing committed authorship never surfaces until the next
+      // edit. Mirrors ThreadController's own registry subscription; `loadAll`
+      // already gates on isCoediting, and exit intentionally does nothing here
+      // (no data to wipe — EditorProposalController's own gate clears the
+      // decorations that read spansFor). Registered before ProposalController/
+      // EditorProposalController (construction order in extension.ts) so their
+      // renders see fresh spans on the same enter transition.
+      this.registry.onDidChange(({ uri, coediting }) => {
+        if (!coediting) {
+          // Finding 5 (final whole-branch review): exit hides the status-bar
+          // item too — otherwise a stale "N orphaned attributions" warning
+          // for a doc that's no longer even coediting stays pinned in the bar
+          // until some unrelated render happens to fire for it.
+          if (vscode.window.activeTextEditor?.document.uri.toString() === uri) this.statusItem.hide();
+          return;
+        }
+        const doc = vscode.workspace.textDocuments.find((d) => d.uri.toString() === uri);
+        if (doc) this.loadAll(doc);
+      }),
    );
  }

  private isTracked(document: vscode.TextDocument): boolean {
-    return document.uri.scheme === "file" && document.uri.fsPath.startsWith(this.rootDir);
+    return isAuthorable(document.uri.scheme);
  }
-  private docPathOf(uri: vscode.Uri): string {
-    return vscode.workspace.asRelativePath(uri, false);
+  /** The single document key (F8): repo-relative path in-workspace, URI string otherwise. */
+  private keyOf(document: vscode.TextDocument): string {
+    return this.store.keyOf(docIdentity(document));
  }
  private currentAuthor(): Provenance {
    const id = vscode.workspace.getConfiguration("git").get<string>("user.name") || process.env.USER || "human";
-    const email = gitUserEmail(this.rootDir);
+    const email = this.rootDir !== undefined ? gitUserEmail(this.rootDir) : undefined;
    return { kind: "human", id, ...(email !== undefined ? { email } : {}) };
  }
  private state(docPath: string): DocAttribution {
@@ -111,7 +128,8 @@ export class AttributionController implements vscode.Disposable {
  /** Load the sidecar and re-resolve every attribution (live span | orphan). */
  loadAll(document: vscode.TextDocument): void {
    if (!this.isTracked(document)) return;
-    const docPath = this.docPathOf(document.uri);
+    if (!this.registry.isCoediting(document.uri)) return;
+    const docPath = this.keyOf(document);
    const s = this.state(docPath);
    const artifact = this.store.load(docPath);
    s.spans = [];
@@ -142,7 +160,7 @@ export class AttributionController implements vscode.Disposable {
  handleExternalSidecarChange(uri: vscode.Uri): void {
    for (const s of this.docs.values()) {
      if (this.store.sidecarPath(s.docPath) === uri.fsPath) {
-        const doc = vscode.workspace.textDocuments.find((d) => this.docPathOf(d.uri) === s.docPath);
+        const doc = vscode.workspace.textDocuments.find((d) => this.keyOf(d) === s.docPath);
        if (doc) this.loadAll(doc);
      }
    }
@@ -152,7 +170,8 @@ export class AttributionController implements vscode.Disposable {

  private onDidChange(e: vscode.TextDocumentChangeEvent): void {
    if (!this.isTracked(e.document) || e.contentChanges.length === 0) return;
-    const docPath = this.docPathOf(e.document.uri);
+    if (!this.registry.isCoediting(e.document.uri)) return;
+    const docPath = this.keyOf(e.document);
    if (!e.document.isDirty && this.matchesDisk(e.document)) {
      // Disk sync (revert / external reload): buffer now equals the file on
      // disk — re-resolve, never attribute (PUC-4). A real edit can also
@@ -162,18 +181,28 @@ export class AttributionController implements vscode.Disposable {
      return;
    }
    const s = this.state(docPath);
+    // An undo/redo is history navigation, NOT authorship (#38): reconcile span
+    // geometry but never freshly attribute the re-inserted text to the current
+    // author — otherwise restored baseline text (or reverted Claude text) is
+    // falsely colored human in the preview. A seam edit is always a forward
+    // apply, so undo/redo also bypasses seam matching.
+    const isUndoRedo =
+      e.reason === vscode.TextDocumentChangeReason.Undo ||
+      e.reason === vscode.TextDocumentChangeReason.Redo;
    // One applyEdit = one change event, but the host may deliver a seam edit
    // as SEVERAL minimal hunks (word-level diffing). Match the EVENT's net
    // effect against the registry; on a hit the agent owns its FULL intended
    // replacement (INV-9) — apply it as ONE algebra edit, not per hunk.
-    const hit = this.pending.matchEvent(
-      docPath,
-      e.contentChanges.map((c) => ({
-        start: c.rangeOffset,
-        end: c.rangeOffset + c.rangeLength,
-        newLength: c.text.length,
-      })),
-    );
+    const hit = isUndoRedo
+      ? null
+      : this.pending.matchEvent(
+          docPath,
+          e.contentChanges.map((c) => ({
+            start: c.rangeOffset,
+            end: c.rangeOffset + c.rangeLength,
+            newLength: c.text.length,
+          })),
+        );
    if (hit) {
      const full = hit.full ?? { start: hit.start, end: hit.end, newLength: hit.newText.length };
      s.spans = applyChange(s.spans, full, hit.provenance, {
@@ -190,10 +219,16 @@ export class AttributionController implements vscode.Disposable {
          end: change.rangeOffset + change.rangeLength,
          newLength: change.text.length,
        };
-        s.spans = applyChange(s.spans, edit, this.currentAuthor(), {
-          newId: () => newId("at"),
-          now: () => new Date().toISOString(),
-        });
+        s.spans = applyChange(
+          s.spans,
+          edit,
+          this.currentAuthor(),
+          {
+            newId: () => newId("at"),
+            now: () => new Date().toISOString(),
+          },
+          !isUndoRedo,
+        );
      }
    }
    if (s.spans.length > 0) s.hadAttributions = true;
@@ -243,11 +278,11 @@ export class AttributionController implements vscode.Disposable {
    range: vscode.Range,
    newText: string,
    provenance: Provenance,
-    opts?: { expectedVersion?: number; turnId?: string },
+    opts?: { expectedVersion?: number; turnId?: string; landBaseline?: boolean },
  ): Promise<boolean> {
    if (!this.isTracked(document)) return false;
    if (opts?.expectedVersion !== undefined && document.version !== opts.expectedVersion) return false;
-    const docPath = this.docPathOf(document.uri);
+    const docPath = this.keyOf(document);
    const startOffset = document.offsetAt(range.start);
    const endOffset = document.offsetAt(range.end);
    const oldText = document.getText(range);
@@ -284,21 +319,34 @@ export class AttributionController implements vscode.Disposable {
        "(host minimized differently?) — the edit may be mis-attributed (INV-9).",
      );
    }
-    if (ok) {
-      // F6 (INV-18): a real machine landing — signal the baseline to advance so
-      // this text never shows as a change in the diff view. Fire regardless of
-      // attribution-match bookkeeping above; the landing happened either way.
+    if (ok && opts?.landBaseline !== false) {
+      // F6 (INV-18, retired as a baseline advance — spec §6.4/INV-7): fire the
+      // machine-landing signal. F12 (INV-48) suppresses this for optimistic
+      // apply: the proposed text is in the buffer but the change stays PENDING
+      // until accept.
      this.applyEmitter.fire({ document });
    }
    return ok;
  }

+  /**
+   * F12/#64 (INV-51): fire the machine-landing signal WITHOUT applying text — used
+   * by finalize-in-place, where the proposed text already landed in the buffer via
+   * optimistic apply (`landBaseline:false`) and accept marks the landing so the
+   * now-accepted change is recorded (INV-7: the baseline itself no longer
+   * advances on landing — it advances only on commit / "Mark Changes as
+   * Reviewed").
+   */
+  signalLanded(document: vscode.TextDocument): void {
+    if (this.isTracked(document)) this.applyEmitter.fire({ document });
+  }
+
  // ---- PUC-4: persistence on save ----------------------------------------------------

  private onDidSave(document: vscode.TextDocument): void {
    if (!this.isTracked(document)) return;
-    if (this.guard.isReadOnly(this.docPathOf(document.uri))) return;
-    const docPath = this.docPathOf(document.uri);
+    if (this.guard.isReadOnly(this.keyOf(document))) return;
+    const docPath = this.keyOf(document);
    const s = this.docs.get(docPath);
    // Allow save when hadAttributions is true even if spans/orphans are now empty:
    // that means the user deliberately deleted all attributed text, and we must
@@ -338,11 +386,6 @@ export class AttributionController implements vscode.Disposable {

  // ---- PUC-5: rendering ----------------------------------------------------------------

-  private toggle(): void {
-    this.visible = !this.visible;
-    this.renderActive();
-  }
-
  private renderActive(): void {
    const editor = vscode.window.activeTextEditor;
    if (editor) this.render(editor.document);
@@ -350,20 +393,7 @@ export class AttributionController implements vscode.Disposable {

  private render(document: vscode.TextDocument): void {
    if (!this.isTracked(document)) return;
-    const s = this.docs.get(this.docPathOf(document.uri));
-    const spans = this.visible && s ? s.spans : [];
-    const toRange = (sp: LiveSpan) =>
-      new vscode.Range(document.positionAt(sp.start), document.positionAt(sp.end));
-    const agentRanges = spans.filter((x) => x.author.kind === "agent").map(toRange);
-    const humanRanges = spans.filter((x) => x.author.kind === "human").map(toRange);
-    // Apply decorations to ALL visible split-editors showing this document, not
-    // just the first match — each editor pane has its own decoration layer.
-    for (const editor of vscode.window.visibleTextEditors) {
-      if (editor.document === document) {
-        editor.setDecorations(this.agentType, agentRanges);
-        editor.setDecorations(this.humanType, humanRanges);
-      }
-    }
+    const s = this.docs.get(this.keyOf(document));
    if (document === vscode.window.activeTextEditor?.document) {
      this.renderStatus(s);
    }
@@ -398,10 +428,19 @@ export class AttributionController implements vscode.Disposable {
  getOrphanCount(docPath: string): number {
    return this.docs.get(docPath)?.orphans.length ?? 0;
  }
-  isVisible(): boolean {
-    return this.visible;
-  }

+  /**
+   * F9: the document's live attribution as authorship spans for the preview —
+   * current-buffer char ranges mapped to author kind (agent→claude). Computes
+   * the document key internally, so callers pass a TextDocument, not the key.
+   */
+  spansFor(document: vscode.TextDocument): AuthorSpan[] {
+    return this.getSpans(this.keyOf(document)).map((s) => ({
+      start: s.range.start,
+      end: s.range.end,
+      author: s.authorKind === "agent" ? "claude" : "human",
+    }));
+  }
  dispose(): void {
    for (const d of this.disposables) d.dispose();
  }
@@ -49,12 +49,19 @@ export function coalesce(spans: LiveSpan[]): LiveSpan[] {
 * Apply one document edit (the half-open range [start,end) replaced by
 * newLength chars) authored by `author`. Existing spans shift/split/clip;
 * inserted chars become a new span of `author` (INV-7).
+ *
+ * `attributeInserted` (default true) controls whether inserted chars get a new
+ * span. Pass `false` for an UNDO/REDO change (#38): the geometry of existing
+ * spans is still reconciled, but re-inserted text is NOT freshly attributed —
+ * an undo is history navigation, not authorship, so restored text stays neutral
+ * rather than being falsely claimed by the current author.
 */
 export function applyChange(
  spans: LiveSpan[],
  edit: TextEdit,
  author: Provenance,
  ctx: TrackerCtx,
+  attributeInserted = true,
 ): LiveSpan[] {
  const delta = edit.newLength - (edit.end - edit.start);
  const out: LiveSpan[] = [];
@@ -71,7 +78,7 @@ export function applyChange(
      if (right) out.push(left ? { ...right, id: ctx.newId() } : right);
    }
  }
-  if (edit.newLength > 0) {
+  if (attributeInserted && edit.newLength > 0) {
    out.push({
      id: ctx.newId(),
      start: edit.start,
@@ -7,11 +7,19 @@
 * key is a hash of the document URI (the controller derives it), not a
 * workspace-relative path; an untitled buffer has no durable identity and is
 * never persisted here (in-memory only — see DiffViewController).
+ *
+ * INV-7 (native-surfaces migration, spec §6.4): the baseline is now either the
+ * git-HEAD blob (never persisted — see GitBaselineAdapter/DiffViewController)
+ * or a snapshot captured on coediting-entry / re-pinned by "Mark Changes as
+ * Reviewed". `"opened"`/`"machine-landing"` are legacy reasons from the
+ * shipped #17/#48 baseline; `normalizeReason` migrates them on load so old
+ * on-disk baselines keep round-tripping under the new vocabulary.
 */
 import * as fs from "node:fs";
 import * as path from "node:path";

-export type BaselineReason = "opened" | "machine-landing" | "pinned";
+export type BaselineReason = "entered" | "pinned" | "head";
+type LegacyBaselineReason = BaselineReason | "opened" | "machine-landing";

 export interface Baseline {
  /** the document URI (the identity the key is derived from) — for round-trip/debug. */
@@ -22,6 +30,12 @@ export interface Baseline {
  reason: BaselineReason;
 }

+/** Migrate legacy on-disk reasons to the current vocabulary (§6.4 INV-7). */
+export function normalizeReason(reason: LegacyBaselineReason): BaselineReason {
+  if (reason === "opened" || reason === "machine-landing") return "entered";
+  return reason;
+}
+
 export class BaselineStore {
  /** @param storageDir absolute VS Code GLOBAL-storage dir (context.globalStorageUri.fsPath). */
  constructor(private readonly storageDir: string) {}
@@ -34,7 +48,8 @@ export class BaselineStore {
  load(key: string): Baseline | null {
    const p = this.baselinePath(key);
    if (!fs.existsSync(p)) return null;
-    return JSON.parse(fs.readFileSync(p, "utf8")) as Baseline;
+    const raw = JSON.parse(fs.readFileSync(p, "utf8")) as Baseline;
+    return { ...raw, reason: normalizeReason(raw.reason) };
  }

  /** Newest epoch wins — overwrite in place, no history (state-not-history, the F3/F4 precedent). */
@@ -0,0 +1,57 @@
+/**
+ * CoeditingRegistry — the INV-10 activation gate (spec §6.4). Holds the set of
+ * documents the writer has explicitly entered into coediting; every surface
+ * (SCM/QuickDiff, comments, preview annotations, edit commands, decorations)
+ * checks membership before attaching. Persisted in workspaceState so a reload
+ * restores the set (the durable sidecar carries the artifacts; this carries
+ * only the mode).
+ */
+import * as vscode from "vscode";
+
+const STATE_KEY = "cowriting.coeditingSet";
+
+export class CoeditingRegistry implements vscode.Disposable {
+  private readonly set: Set<string>;
+  private readonly emitter = new vscode.EventEmitter<{ uri: string; coediting: boolean }>();
+  readonly onDidChange = this.emitter.event;
+  /** Set by the SCM surface (Task 3) so syncContext can expose the baseline mode. */
+  baselineModeOf: ((uri: vscode.Uri) => "head" | "snapshot" | undefined) | undefined;
+
+  constructor(private readonly state: vscode.Memento) {
+    this.set = new Set(state.get<string[]>(STATE_KEY, []));
+  }
+
+  enter(uri: vscode.Uri): void {
+    const key = uri.toString();
+    if (this.set.has(key)) return;
+    this.set.add(key);
+    void this.state.update(STATE_KEY, [...this.set]);
+    this.emitter.fire({ uri: key, coediting: true });
+  }
+
+  exit(uri: vscode.Uri): void {
+    const key = uri.toString();
+    if (!this.set.delete(key)) return;
+    void this.state.update(STATE_KEY, [...this.set]);
+    this.emitter.fire({ uri: key, coediting: false });
+  }
+
+  isCoediting(uri: vscode.Uri): boolean {
+    return this.set.has(uri.toString());
+  }
+
+  list(): string[] {
+    return [...this.set];
+  }
+
+  syncContext(editor: vscode.TextEditor | undefined): void {
+    const coediting = !!editor && this.isCoediting(editor.document.uri);
+    void vscode.commands.executeCommand("setContext", "cowriting.isCoediting", coediting);
+    const mode = coediting && editor ? this.baselineModeOf?.(editor.document.uri) : undefined;
+    void vscode.commands.executeCommand("setContext", "cowriting.baselineMode", mode ?? "");
+  }
+
+  dispose(): void {
+    this.emitter.dispose();
+  }
+}
@@ -1,64 +1,84 @@
 /**
- * DiffViewController — F6 diff-view toggle (spec §6.2/§6.4). Owns the baseline
- * lifecycle (initialize at first sight / advance at every machine landing / pin
- * on demand), serves the baseline as a readonly `cowriting-baseline:` virtual
- * document, and toggles a native vscode.diff (baseline left, the LIVE document
- * right). A pure view: never mutates the document, sidecar, or attribution
- * state (INV-19).
+ * DiffViewController — the baseline router (native-surfaces migration, spec
+ * §6.4, INV-7). Resolves each coedited document's review baseline to one of two
+ * modes: **head** (git-tracked — the baseline is the `HEAD` blob, re-read on
+ * every commit, never persisted) or **snapshot** (untracked/no-repo — captured
+ * on coediting-entry, re-pinned on demand by "Mark Changes as Reviewed").
+ * Baselines are established ONLY for documents the CoeditingRegistry has
+ * entered (INV-10) — opening a file no longer captures one (the retired
+ * `ensureBaseline`/#17 behavior). The shipped machine-landing advance (#48,
+ * INV-18) is retired too: a landed Claude edit stays a visible change-since-
+ * baseline until the file is committed (head mode) or reviewed (snapshot
+ * mode) — spec INV-7's note, D21. A pure data layer: never mutates the
+ * document, sidecar, or attribution state (INV-19).
 *
- * F6 works on ANY text document, not just workspace files — it needs no
+ * The F6 two-pane `vscode.diff` *view* (toggle UI + the `cowriting-baseline:`
+ * virtual document) was removed in #34 once F10 made the rendered preview the
+ * single review surface; only the baseline store survives here (spec §6.7).
+ *
+ * The baseline works on ANY text document, not just workspace files — it needs no
 * `.threads/` sidecar, only a stable doc identity + a storage home. So it has
 * its OWN diffable predicate (any `file:` or `untitled:` doc), decoupled from
- * F2's workspace `isTracked`. Persistable docs (`file:`) keep their baseline in
- * VS Code's per-extension GLOBAL storage keyed by a hash of the document URI;
- * untitled buffers have no durable identity, so their baseline is in-memory
- * only (lost on reload) — the same degrade the storage-unavailable path uses.
+ * F2's workspace `isTracked`. Persistable docs (`file:`) keep their SNAPSHOT
+ * baseline in VS Code's per-extension GLOBAL storage keyed by a hash of the
+ * document URI (INV-7: head-mode baselines are storage-free — never
+ * persisted); untitled buffers have no durable identity, so their baseline is
+ * in-memory only (lost on reload) — the same degrade the storage-unavailable
+ * path uses.
 */
-import * as path from "node:path";
 import { createHash } from "node:crypto";
 import * as vscode from "vscode";
 import { BaselineStore, type Baseline, type BaselineReason } from "./baselineStore";
-
-export const BASELINE_SCHEME = "cowriting-baseline";
+import type { GitBaselineAdapter } from "./gitBaseline";
+import type { CoeditingRegistry } from "./coeditingRegistry";

 export class DiffViewController implements vscode.Disposable {
  private readonly disposables: vscode.Disposable[] = [];
-  /** Source of truth for the content provider, keyed by `document.uri.toString()`. */
+  /** Source of truth for the baseline, keyed by `document.uri.toString()`. */
  private readonly baselines = new Map<string, Baseline>();
-  private readonly onDidChangeEmitter = new vscode.EventEmitter<vscode.Uri>();
+  /** Which mode each established document resolved to (INV-7). */
+  private readonly modes = new Map<string, "head" | "snapshot">();
+  /** F7 (additive): fires on every baseline capture (establish / head-refresh /
+   *  pin) so the track-changes preview refreshes without polling. Carries the
+   *  real document URI. */
+  private readonly onDidChangeBaselineEmitter = new vscode.EventEmitter<{ uri: string }>();
+  readonly onDidChangeBaseline = this.onDidChangeBaselineEmitter.event;
  private storageWarned = false;

-  constructor(private readonly store: BaselineStore | null) {
-    const provider: vscode.TextDocumentContentProvider = {
-      onDidChange: this.onDidChangeEmitter.event,
-      provideTextDocumentContent: (uri) => {
-        // The baseline URI carries the real document URI in its query (see baselineUri).
-        return this.baselines.get(uri.query)?.text ?? "";
-      },
-    };
+  constructor(
+    private readonly store: BaselineStore | null,
+    private readonly git: GitBaselineAdapter,
+    registry: CoeditingRegistry,
+  ) {
    this.disposables.push(
-      this.onDidChangeEmitter,
-      vscode.workspace.registerTextDocumentContentProvider(BASELINE_SCHEME, provider),
-      vscode.commands.registerCommand("cowriting.toggleDiffView", () =>
-        this.toggle(vscode.window.activeTextEditor),
-      ),
-      vscode.commands.registerCommand("cowriting.pinDiffBaseline", () =>
+      this.onDidChangeBaselineEmitter,
+      vscode.commands.registerCommand("cowriting.markReviewed", () =>
        this.pinCommand(vscode.window.activeTextEditor),
      ),
-      // F6 captures a baseline for any diffable doc the moment it is first seen,
-      // independent of the workspace gate (so "opened" is the open-time text).
-      vscode.workspace.onDidOpenTextDocument((d) => this.ensureBaseline(d)),
+      // INV-10: a baseline is established only once its document ENTERS
+      // coediting; exiting keeps the stored baseline as-is (PUC-7).
+      registry.onDidChange(({ uri, coediting }) => {
+        if (!coediting) return;
+        const doc = vscode.workspace.textDocuments.find((d) => d.uri.toString() === uri);
+        if (doc) void this.establish(doc);
+      }),
+      // A commit on a watched repo may advance a head-mode baseline (D13).
+      git.onDidChangeHead(() => void this.refreshHeadBaselines()),
    );
-    for (const d of vscode.workspace.textDocuments) this.ensureBaseline(d);
+    // Restore already-coediting documents (e.g. after an extension host reload).
+    for (const key of registry.list()) {
+      const doc = vscode.workspace.textDocuments.find((d) => d.uri.toString() === key);
+      if (doc) void this.establish(doc);
+    }
  }

  // ---- diffability / identity --------------------------------------------------------

-  /** Any text document F6 can diff: a saved file OR an unsaved buffer. */
+  /** Any text document F6 tracks a baseline for: a saved file OR an unsaved buffer. */
  private isDiffable(document: vscode.TextDocument): boolean {
    return document.uri.scheme === "file" || document.uri.scheme === "untitled";
  }
-  /** Only `file:` docs have a durable identity to persist a baseline against. */
+  /** Only `file:` docs have a durable identity to persist a SNAPSHOT baseline against. */
  private isPersistable(document: vscode.TextDocument): boolean {
    return document.uri.scheme === "file";
  }
@@ -70,137 +90,128 @@ export class DiffViewController implements vscode.Disposable {
  private storageKey(uriKey: string): string {
    return createHash("sha256").update(uriKey).digest("hex");
  }
-  /** The readonly virtual-doc URI; its query carries the real document URI. */
-  private baselineUri(document: vscode.TextDocument): vscode.Uri {
-    const name = path.basename(document.uri.path) || "untitled";
-    return vscode.Uri.from({ scheme: BASELINE_SCHEME, path: "/" + name, query: this.uriKey(document) });
-  }

-  // ---- baseline lifecycle (§6.4) -----------------------------------------------------
+  // ---- baseline lifecycle (§6.4, INV-7) -----------------------------------------------

-  /** First sight of a diffable doc: load the stored baseline, else capture `opened`. */
-  ensureBaseline(document: vscode.TextDocument): void {
+  /**
+   * Resolve this document's baseline mode and (re)establish its baseline.
+   * git-tracked (a HEAD blob resolves) → **head** mode, baseline = HEAD, never
+   * persisted. Otherwise → **snapshot** mode: load a previously-stored
+   * snapshot, else capture the buffer now (`"entered"`). Called once when a
+   * document enters coediting (registry.enter) and at startup for documents
+   * already coediting.
+   */
+  async establish(document: vscode.TextDocument): Promise<void> {
    if (!this.isDiffable(document)) return;
    const key = this.uriKey(document);
-    if (this.baselines.has(key)) return;
+    const head = await this.git.headFor(document.uri);
+    if (head) {
+      this.modes.set(key, "head");
+      this.setBaseline(key, head.text, "head");
+      return;
+    }
+    this.modes.set(key, "snapshot");
    if (this.store && this.isPersistable(document)) {
-      try {
-        const stored = this.store.load(this.storageKey(key));
-        if (stored) {
-          this.baselines.set(key, stored);
-          return;
-        }
-      } catch {
-        this.warnStorageOnce();
+      const stored = this.tryLoad(key); // existing load path, legacy reasons normalized
+      if (stored) {
+        this.baselines.set(key, stored);
+        this.fire(key);
+        return;
      }
    }
-    this.capture(document, "opened");
+    this.capture(document, "entered");
  }

-  /** Machine landing (INV-18): re-capture so landed text never shows as a change. */
-  advance(document: vscode.TextDocument): void {
-    if (!this.isDiffable(document)) return;
-    this.capture(document, "machine-landing");
+  /** The mode a document's baseline resolved to, or undefined before `establish`. */
+  modeOf(uriString: string): "head" | "snapshot" | undefined {
+    return this.modes.get(uriString);
  }

-  /** Human pin: baseline := now; the open diff visibly empties (left = right). */
+  /** Re-read HEAD for every head-mode document; a commit advances the baseline (D13). */
+  private async refreshHeadBaselines(): Promise<void> {
+    for (const [key, mode] of this.modes) {
+      if (mode !== "head") continue;
+      const doc = vscode.workspace.textDocuments.find((d) => d.uri.toString() === key);
+      if (!doc) continue;
+      const head = await this.git.headFor(doc.uri);
+      if (head && head.text !== this.baselines.get(key)?.text) {
+        this.setBaseline(key, head.text, "head"); // commit advanced the baseline (D13)
+      }
+    }
+  }
+
+  /** Human pin ("Mark Changes as Reviewed", D14): snapshot mode only — a
+   *  git-tracked (head-mode) baseline advances only by commit. */
  pin(document: vscode.TextDocument): void {
    if (!this.isDiffable(document)) return;
+    const key = this.uriKey(document);
+    const mode = this.modes.get(key);
+    if (mode === undefined) {
+      // never established — the document is not being coedited (#71 item 1)
+      void vscode.window.showWarningMessage("Run ✦ Coedit this Document with Claude first.");
+      return;
+    }
+    if (mode !== "snapshot") {
+      void vscode.window.showWarningMessage(
+        "Cowriting: this document is git-tracked — commit to advance the baseline.",
+      );
+      return;
+    }
    this.capture(document, "pinned");
  }

-  /** Capture buffer text at this epoch, persist (if persistable), refresh any open diff. */
+  /** Capture buffer text at this epoch, persist (if persistable), notify F7/F10. */
  private capture(document: vscode.TextDocument, reason: BaselineReason): void {
    const key = this.uriKey(document);
    const baseline: Baseline = { uri: key, text: document.getText(), capturedAt: new Date().toISOString(), reason };
    this.baselines.set(key, baseline);
    if (this.store && this.isPersistable(document)) {
-      try {
-        this.store.save(this.storageKey(key), baseline);
-      } catch {
-        this.warnStorageOnce();
-      }
+      this.trySave(key, baseline);
    }
-    // An open diff re-requests the left side when its baseline URI changes.
-    this.onDidChangeEmitter.fire(this.baselineUri(document));
+    this.fire(key);
+  }
+
+  /** Set a baseline from resolved text directly (head mode — no document buffer read). */
+  private setBaseline(key: string, text: string, reason: BaselineReason): void {
+    const baseline: Baseline = { uri: key, text, capturedAt: new Date().toISOString(), reason };
+    this.baselines.set(key, baseline);
+    // head-mode baselines are storage-free (INV-7); persist only snapshots.
+    if (reason !== "head" && this.store) this.trySave(key, baseline);
+    this.fire(key);
+  }
+
+  private tryLoad(key: string): Baseline | null {
+    try {
+      return this.store!.load(this.storageKey(key));
+    } catch {
+      this.warnStorageOnce();
+      return null;
+    }
+  }
+
+  private trySave(key: string, baseline: Baseline): void {
+    try {
+      this.store!.save(this.storageKey(key), baseline);
+    } catch {
+      this.warnStorageOnce();
+    }
+  }
+
+  private fire(key: string): void {
+    this.onDidChangeBaselineEmitter.fire({ uri: key });
  }

  private warnStorageOnce(): void {
    if (this.storageWarned) return;
    this.storageWarned = true;
    void vscode.window.showWarningMessage(
-      "Cowriting: diff-view storage is unavailable — baselines are kept in memory only and won't survive a reload.",
+      "Cowriting: baseline storage is unavailable — baselines are kept in memory only and won't survive a reload.",
    );
  }

-  // ---- toggle UX (§6.5 PUC-1) --------------------------------------------------------
-
-  /**
-   * If this doc's baseline diff is open → close it and reveal the normal editor;
-   * else open vscode.diff (baseline left, the live document right). Warns only
-   * when there is no active text editor (or a non-diffable scheme).
-   */
-  private async toggle(editor: vscode.TextEditor | undefined): Promise<void> {
-    if (!editor || !this.isDiffable(editor.document)) {
-      void vscode.window.showWarningMessage(
-        "Cowriting: focus a text editor to toggle its diff view.",
-      );
-      return;
-    }
-    const document = editor.document;
-    const openTab = this.findDiffTab(document.uri);
-    if (openTab) {
-      await vscode.window.tabGroups.close(openTab);
-      await vscode.window.showTextDocument(document, { preview: false });
-      return;
-    }
-    this.ensureBaseline(document);
-    const baseline = this.baselines.get(this.uriKey(document))!;
-    const name = path.basename(document.uri.path) || "untitled";
-    const unsaved = this.isPersistable(document) ? "" : " (unsaved)";
-    const title = `${name} — my changes since ${this.epochLabel(baseline)}${unsaved}`;
-    await vscode.commands.executeCommand(
-      "vscode.diff",
-      this.baselineUri(document),
-      document.uri,
-      title,
-      { preview: false },
-    );
-  }
-
-  /** The open baseline-diff tab for this document, if any. */
-  private findDiffTab(modified: vscode.Uri): vscode.Tab | undefined {
-    for (const group of vscode.window.tabGroups.all) {
-      for (const tab of group.tabs) {
-        const input = tab.input;
-        if (
-          input instanceof vscode.TabInputTextDiff &&
-          input.original.scheme === BASELINE_SCHEME &&
-          input.modified.toString() === modified.toString()
-        ) {
-          return tab;
-        }
-      }
-    }
-    return undefined;
-  }
-
-  /** Human-readable epoch for the diff tab title (§5 / §6.5). */
-  private epochLabel(baseline: Baseline): string {
-    const time = new Date(baseline.capturedAt).toLocaleTimeString();
-    switch (baseline.reason) {
-      case "opened":
-        return `opened ${time}`;
-      case "machine-landing":
-        return `Claude landed ${time}`;
-      case "pinned":
-        return `pinned ${time}`;
-    }
-  }
-
  private pinCommand(editor: vscode.TextEditor | undefined): void {
    if (!editor || !this.isDiffable(editor.document)) {
-      void vscode.window.showWarningMessage("Cowriting: focus a text editor to pin its diff baseline.");
+      void vscode.window.showWarningMessage("Cowriting: focus a text editor to mark its changes as reviewed.");
      return;
    }
    this.pin(editor.document);
@@ -212,10 +223,6 @@ export class DiffViewController implements vscode.Disposable {
    const b = this.baselines.get(uriString);
    return b ? { text: b.text, reason: b.reason, capturedAt: b.capturedAt } : undefined;
  }
-  /** Is this doc's baseline diff currently open in any tab group? */
-  isDiffOpen(uriString: string): boolean {
-    return this.findDiffTab(vscode.Uri.parse(uriString)) !== undefined;
-  }
  /** Absolute on-disk path of this doc's persisted baseline, or undefined (untitled/in-memory). */
  baselineFilePath(uriString: string): string | undefined {
    if (!this.store || vscode.Uri.parse(uriString).scheme !== "file") return undefined;
@@ -0,0 +1,102 @@
+/**
+ * EditFlow — the host edit-turn flow (F11/F12), extracted from the review
+ * webview controller ahead of its sunset (native-surfaces migration, spec
+ * §6.10). Owns the low-level turn→proposal(s) cut (`runEditAndPropose`,
+ * INV-39/40): never mutates the document directly — every turn lands as one or
+ * more F4 proposals via `ProposalController.propose` (INV-10). Reachable, from
+ * Task 6 on, from the thread controller (`ThreadController.runMakeEdit`) —
+ * vscode-API-only, no webview state. The `cowriting.editDocument` command lives
+ * in extension.ts (Finding-1 fix) so it can hand off to
+ * `ThreadController.askClaude()`. Task 8 deleted this class's own
+ * `askClaude`/`askEditInstruction` (the review webview's toolbar ask) once
+ * their only caller — the webview — was sunset alongside them.
+ */
+import * as vscode from "vscode";
+import type { ProposalController } from "./proposalController";
+import { diffToBlockHunks } from "./trackChangesModel";
+import { buildFingerprint } from "./anchorer";
+import type { EditTurnResult, RunEditTurnOptions } from "./liveTurn";
+
+/**
+ * F11: a host edit turn (selection/document text + instruction → rewrite).
+ * Injectable for tests. #60: accepts optional turn options (onProgress/signal);
+ * the arg is optional so existing test stubs that ignore it stay valid.
+ */
+type EditTurn = (instruction: string, text: string, opts?: RunEditTurnOptions) => Promise<EditTurnResult>;
+/** F11: what an Ask-Claude gesture edits — a resolved selection range, or the whole document. */
+export type EditTarget = { kind: "range"; start: number; end: number } | { kind: "document" };
+
+export class EditFlow implements vscode.Disposable {
+  private readonly disposables: vscode.Disposable[] = [];
+  /**
+   * F11: the host edit turn (INV-8 — runs host-side, @cline/sdk loaded lazily and
+   * never bundled). Injectable so host E2E can stub it (no LLM in CI).
+   */
+  private editTurn: EditTurn = async (instruction, text, opts) => {
+    const { runEditTurn } = await import("./liveTurn");
+    return runEditTurn(instruction, text, opts);
+  };
+  /** Monotonic per-session counter minting a stable turnId for each Ask-Claude gesture. */
+  private turnSeq = 0;
+  private nextTurnSeq(): number {
+    return ++this.turnSeq;
+  }
+
+  constructor(private readonly proposals: ProposalController) {}
+
+  /**
+   * F11/F12 (INV-35/39): run one host edit turn and record the result as F4
+   * proposal(s) — a SELECTION yields one single-range proposal over the resolved
+   * block-union; a DOCUMENT rewrite is `diffToBlockHunks`'d into one proposal per
+   * changed BLOCK (#47, INV-39 supersedes INV-37's per-word cut), each tagged
+   * `granularity:"block"` so accept reconciles attribution per word (INV-40).
+   * Never mutates the document (INV-10). Returns the created proposal ids.
+   */
+  async runEditAndPropose(
+    document: vscode.TextDocument,
+    target: EditTarget,
+    instruction: string,
+    opts?: RunEditTurnOptions,
+  ): Promise<string[]> {
+    const full = document.getText();
+    // One turnId per gesture — the document case's N hunk-proposals all share it,
+    // so a single rewrite groups as one agent turn (parity with editSelection).
+    const turnId = `turn-${this.nextTurnSeq()}`;
+    const provenance = (turn: EditTurnResult) =>
+      ({ kind: "agent" as const, id: "claude", agent: { sdk: "@cline/sdk", model: turn.model, sessionId: turn.sessionId } });
+    if (target.kind === "range") {
+      const selected = full.slice(target.start, target.end);
+      const turn = await this.editTurn(instruction, selected, opts);
+      if (turn.replacement === "" || turn.replacement === selected) return [];
+      const fp = buildFingerprint(full, { start: target.start, end: target.end });
+      const id = await this.proposals.propose(document, fp, turn.replacement, provenance(turn), { turnId, instruction });
+      return id ? [id] : [];
+    }
+    const turn = await this.editTurn(instruction, full, opts);
+    const ids: string[] = [];
+    // #47 (INV-39, supersedes INV-37): a document rewrite is cut at BLOCK
+    // granularity — one proposal per changed block (the unit a human reviews) —
+    // not per word. Each is tagged `granularity:"block"` so accept reconciles
+    // attribution per word inside the block (INV-40).
+    for (const h of diffToBlockHunks(full, turn.replacement)) {
+      const fp = buildFingerprint(full, { start: h.start, end: h.end });
+      const id = await this.proposals.propose(document, fp, h.replacement, provenance(turn), {
+        turnId,
+        instruction,
+        granularity: "block",
+      });
+      if (id) ids.push(id);
+    }
+    return ids;
+  }
+
+  // ---- test seam (§6.4) ----
+  /** F11 test seam: stub the host edit turn so the document/selection paths run without an LLM. */
+  setEditTurnForTest(fn: EditTurn): void {
+    this.editTurn = fn;
+  }
+
+  dispose(): void {
+    for (const d of this.disposables) d.dispose();
+  }
+}
@@ -0,0 +1,288 @@
+/**
+ * EditorProposalController — F12/#64 editor surface (spec coauthoring-inline-editor-diff
+ * §3.5). Reverses INV-32 for pending proposals: on `onDidChangeProposals` it
+ * (1) optimistically applies any not-yet-applied proposal into the active editor's
+ * buffer (ProposalController.optimisticApply, INV-48), (2) decorates each applied
+ * proposal — insertion tint over the proposed text + a non-editable struck-red hint
+ * for deletions (INV-52, decorationPlan/INV-49), and (3) provides a CodeLens
+ * `Accept ▾ / Reject ▾` above each block whose ▾ opens a QuickPick (this / all).
+ * Owns no proposal STATE — it is a view over ProposalController (which stays the
+ * pure F4 owner). Phase 2 decorates ALL committed changes-since-baseline by author
+ * (human green / Claude blue + struck hints for deletions), plus pending proposals;
+ * a pinned baseline suppresses committed marks entirely (pin→clean, INV-48).
+ */
+import * as vscode from "vscode";
+import type { ProposalController } from "./proposalController";
+import type { DiffViewController } from "./diffViewController";
+import type { AttributionController } from "./attributionController";
+import { decorationPlan, wordEditHunks, authorAt } from "./trackChangesModel";
+import { isAuthorable } from "./workspacePath";
+import type { CoeditingRegistry } from "./coeditingRegistry";
+
+export class EditorProposalController implements vscode.Disposable, vscode.CodeLensProvider {
+  private readonly disposables: vscode.Disposable[] = [];
+  private readonly insDeco: Record<"human" | "claude", vscode.TextEditorDecorationType> = {
+    human: vscode.window.createTextEditorDecorationType({
+      backgroundColor: "rgba(63,185,80,0.14)", borderColor: "#3fb950",
+      border: "0 0 2px 0", borderStyle: "solid",
+    }),
+    claude: vscode.window.createTextEditorDecorationType({
+      backgroundColor: "rgba(88,166,255,0.15)", borderColor: "#58a6ff",
+      border: "0 0 2px 0", borderStyle: "solid",
+    }),
+  };
+  private readonly delDeco: Record<"human" | "claude" | "none", vscode.TextEditorDecorationType> = {
+    human: vscode.window.createTextEditorDecorationType({ after: { color: "#f85149" } }),
+    claude: vscode.window.createTextEditorDecorationType({ after: { color: "#bc8cff" } }),
+    none: vscode.window.createTextEditorDecorationType({ after: { color: new vscode.ThemeColor("descriptionForeground") } }),
+  };
+  private readonly lensEmitter = new vscode.EventEmitter<void>();
+  readonly onDidChangeCodeLenses = this.lensEmitter.event;
+  /** Pending debounce timers — one per URI — coalesce rapid-fire propose events
+   * (e.g. runEditAndPropose's N sequential propose() calls) into a single
+   * optimistic-apply pass that runs after ALL proposals are created. Without this,
+   * each propose() fires onDidChangeProposals synchronously and the controller's
+   * optimisticApply runs concurrently with the still-in-progress propose loop,
+   * causing "file changed in the meantime" workspace-edit conflicts. */
+  private readonly pendingApply = new Map<string, ReturnType<typeof setTimeout>>();
+  /** Debounce timers for committed-change re-render (one per URI, 50 ms). */
+  private readonly pendingRender = new Map<string, ReturnType<typeof setTimeout>>();
+
+  constructor(
+    private readonly proposals: ProposalController,
+    private readonly diffView: DiffViewController,
+    private readonly attribution: AttributionController,
+    private readonly registry: CoeditingRegistry,
+  ) {
+    this.disposables.push(
+      this.insDeco.human, this.insDeco.claude, this.delDeco.human, this.delDeco.claude, this.delDeco.none, this.lensEmitter,
+      vscode.languages.registerCodeLensProvider({ language: "markdown" }, this),
+      this.proposals.onDidChangeProposals(({ uri }) => this.scheduleApply(uri)),
+      vscode.window.onDidChangeActiveTextEditor((ed) => ed && this.renderEditor(ed)),
+      // Refresh committed decorations when the baseline changes (establish / pin / head-refresh).
+      this.diffView.onDidChangeBaseline(({ uri }) => {
+        const ed = vscode.window.visibleTextEditors.find((e) => e.document.uri.toString() === uri);
+        if (ed) this.renderEditor(ed);
+      }),
+      // Refresh committed decorations (debounced) when the active doc changes — attribution spans shift.
+      vscode.workspace.onDidChangeTextDocument((e) => {
+        const ed = vscode.window.activeTextEditor;
+        if (ed && e.document === ed.document) this.scheduleRender(ed);
+      }),
+      // INV-10: re-render (and refresh CodeLens) on every gate change — exit clears
+      // decorations/CodeLens via renderEditor's own gate check, re-entry redraws.
+      this.registry.onDidChange(({ uri }) => {
+        const ed = vscode.window.visibleTextEditors.find((e) => e.document.uri.toString() === uri);
+        if (ed) this.renderEditor(ed);
+        this.lensEmitter.fire();
+      }),
+      // the four QuickPick-backed menu commands
+      vscode.commands.registerCommand("cowriting.proposalAcceptMenu", (id?: string) => this.menu("accept", id)),
+      vscode.commands.registerCommand("cowriting.proposalRejectMenu", (id?: string) => this.menu("reject", id)),
+    );
+  }
+
+  /** Debounce-schedule an optimistic-apply pass for the given URI. Multiple rapid
+   * onDidChangeProposals events (from a single runEditAndPropose batch) collapse
+   * into one pass that runs after the batch completes. */
+  private scheduleApply(uri: string): void {
+    const prev = this.pendingApply.get(uri);
+    if (prev !== undefined) clearTimeout(prev);
+    this.pendingApply.set(
+      uri,
+      setTimeout(() => {
+        this.pendingApply.delete(uri);
+        void this.onProposalsChanged(uri);
+      }, 0),
+    );
+  }
+
+  /** Debounce a committed-change re-render for `editor` (50 ms — lets rapid typing settle). */
+  private scheduleRender(editor: vscode.TextEditor): void {
+    const uri = editor.document.uri.toString();
+    const prev = this.pendingRender.get(uri);
+    if (prev !== undefined) clearTimeout(prev);
+    this.pendingRender.set(
+      uri,
+      setTimeout(() => {
+        this.pendingRender.delete(uri);
+        this.renderEditor(editor);
+      }, 50),
+    );
+  }
+
+  /** Apply any not-yet-applied proposals on the doc, then re-decorate + refresh lenses. */
+  private async onProposalsChanged(uri: string): Promise<void> {
+    const doc = vscode.workspace.textDocuments.find((d) => d.uri.toString() === uri);
+    if (!doc || doc.languageId !== "markdown" || !isAuthorable(doc.uri.scheme)) return;
+    // INV-10: a doc that exited coediting between the schedule and this tick gets
+    // no optimistic-apply/decorations/CodeLens.
+    if (!this.registry.isCoediting(doc.uri)) return;
+    const key = this.proposals.keyFor(doc);
+    for (const v of this.proposals.listProposals(doc)) {
+      if (v.anchorStart !== null && !this.proposals.isApplied(key, v.id)) {
+        await this.proposals.optimisticApply(doc, v.id); // fires onDidChangeProposals again; guarded by isApplied
+      }
+    }
+    const ed = vscode.window.visibleTextEditors.find((e) => e.document.uri.toString() === uri);
+    if (ed) this.renderEditor(ed);
+    this.lensEmitter.fire();
+  }
+
+  /** Decorate the editor for every applied proposal on its document (INV-52). */
+  private renderEditor(editor: vscode.TextEditor): void {
+    const doc = editor.document;
+    const clearAll = () => {
+      for (const a of ["human", "claude"] as const) editor.setDecorations(this.insDeco[a], []);
+      for (const a of ["human", "claude", "none"] as const) editor.setDecorations(this.delDeco[a], []);
+    };
+    if (doc.languageId !== "markdown") return clearAll();
+    // INV-10: a non-entered doc gets no decorations at all.
+    if (!this.registry.isCoediting(doc.uri)) return clearAll();
+    const key = this.proposals.keyFor(doc);
+    const ins: Record<"human" | "claude", vscode.Range[]> = { human: [], claude: [] };
+    const del: Record<"human" | "claude" | "none", vscode.DecorationOptions[]> = { human: [], claude: [], none: [] };
+    // Pending proposals — always Claude (INV: proposals are agent-authored).
+    for (const v of this.proposals.listProposals(doc)) {
+      if (v.anchorStart === null || v.original === undefined || !this.proposals.isApplied(key, v.id)) continue;
+      const plan = decorationPlan(v.anchorStart, v.original, v.replacement);
+      for (const i of plan.insertions) ins.claude.push(new vscode.Range(doc.positionAt(i.start), doc.positionAt(i.end)));
+      for (const d of plan.deletions) {
+        del.claude.push({
+          range: new vscode.Range(doc.positionAt(d.at), doc.positionAt(d.at)),
+          renderOptions: { after: { contentText: ` ${d.text} `, textDecoration: "line-through" } },
+        });
+      }
+    }
+    // Committed changes-since-baseline are added by decorateCommitted (pushes into ins/del[author]).
+    this.decorateCommitted(editor, ins, del);
+    for (const a of ["human", "claude"] as const) editor.setDecorations(this.insDeco[a], ins[a]);
+    for (const a of ["human", "claude", "none"] as const) editor.setDecorations(this.delDeco[a], del[a]);
+  }
+
+  /**
+   * Decorates committed (non-proposal) changes-since-baseline by author (Task 7 / reverse INV-32).
+   * Diffs current buffer against the F6 baseline; attributes each changed run to its author;
+   * skips any hunk overlapping an applied pending proposal (the proposal loop owns those).
+   * pin→clean: returns without marking when baseline.reason === "pinned".
+   * Adjacency heuristic: a deletion in a hunk that has insertions inherits the hunk author;
+   * a standalone deletion (hunk with no insertions) renders neutral ("none") — matching the
+   * preview's cw-del-none treatment.
+   */
+  private decorateCommitted(
+    editor: vscode.TextEditor,
+    ins: Record<"human" | "claude", vscode.Range[]>,
+    del: Record<"human" | "claude" | "none", vscode.DecorationOptions[]>,
+  ): void {
+    const doc = editor.document;
+    const baseline = this.diffView.getBaseline(doc.uri.toString());
+    // No baseline yet, or baseline was pinned (pin→clean: no committed marks, INV-48).
+    if (!baseline || baseline.reason === "pinned") return;
+    const current = doc.getText();
+    // Nothing changed — short-circuit before the diff.
+    if (current === baseline.text) return;
+    const spans = this.attribution.spansFor(doc);
+    const key = this.proposals.keyFor(doc);
+    // Build the set of current-buffer ranges owned by applied pending proposals so we
+    // can skip hunks that fall inside them — the proposal loop already decorates those.
+    const appliedRanges: { start: number; end: number }[] = [];
+    for (const v of this.proposals.listProposals(doc)) {
+      if (v.anchorStart !== null && v.original !== undefined && this.proposals.isApplied(key, v.id)) {
+        appliedRanges.push({ start: v.anchorStart, end: v.anchorStart + v.replacement.length });
+      }
+    }
+    // wordEditHunks(current, baseline.text): start/end are in current coords; replacement is baseline text.
+    const hunks = wordEditHunks(current, baseline.text);
+    for (const h of hunks) {
+      // Skip hunks whose current-buffer range overlaps an applied pending proposal.
+      // The skip is whole-hunk and therefore conservative: a committed edit word-merged
+      // into a proposal's hunk is skipped entirely — rare at word granularity.
+      if (appliedRanges.some((r) => h.start < r.end && h.end > r.start)) continue;
+      const author = authorAt(h.start, spans) ?? "human";
+      // h.replacement = baseline text for this span; current.slice(h.start, h.end) = applied text.
+      const plan = decorationPlan(h.start, h.replacement, current.slice(h.start, h.end));
+      for (const i of plan.insertions) {
+        ins[author].push(new vscode.Range(doc.positionAt(i.start), doc.positionAt(i.end)));
+      }
+      // Adjacency heuristic: a deletion paired with an insertion in THIS hunk inherits the
+      // insertion author; a standalone deletion (no insertion in the hunk) is neutral.
+      const delAuthor = plan.insertions.length > 0 ? author : "none";
+      for (const d of plan.deletions) {
+        del[delAuthor].push({
+          range: new vscode.Range(doc.positionAt(d.at), doc.positionAt(d.at)),
+          renderOptions: { after: { contentText: ` ${d.text} `, textDecoration: "line-through" } },
+        });
+      }
+    }
+  }
+
+  /**
+   * CodeLensProvider (Task 8, PUC-2 wording): a `✓ Keep` / `✗ Reject` pair above
+   * each applied block, plus — when ≥2 proposals are pending on the doc — a
+   * top-of-file `✓ Keep all (N)` / `✗ Reject all` pair routed directly at the
+   * batch commands (no QuickPick detour for the common "review is done" case).
+   */
+  provideCodeLenses(document: vscode.TextDocument): vscode.CodeLens[] {
+    if (document.languageId !== "markdown") return [];
+    if (!this.registry.isCoediting(document.uri)) return [];
+    const key = this.proposals.keyFor(document);
+    // #70: an interior tweak orphans the exact anchor (anchorStart null) but the
+    // proposal stays fully decidable — Keep by id, Reject via the tracked span —
+    // so the lens pair anchors at the tracked span instead of vanishing (which
+    // made the fixed reject path unreachable from the primary gesture surface).
+    const applied: { id: string; at: number }[] = [];
+    for (const v of this.proposals.listProposals(document)) {
+      if (!this.proposals.isApplied(key, v.id)) continue;
+      const at = v.anchorStart ?? this.proposals.trackedSpan(key, v.id)?.start;
+      if (at !== undefined) applied.push({ id: v.id, at });
+    }
+    const lenses: vscode.CodeLens[] = [];
+    if (applied.length >= 2) {
+      const top = new vscode.Range(0, 0, 0, 0);
+      lenses.push(
+        new vscode.CodeLens(top, { title: `✓ Keep all (${applied.length})`, command: "cowriting.acceptAllProposals" }),
+        new vscode.CodeLens(top, { title: "✗ Reject all", command: "cowriting.rejectAllProposals" }),
+      );
+    }
+    for (const v of applied) {
+      const pos = document.positionAt(v.at);
+      const line = new vscode.Range(pos.line, 0, pos.line, 0);
+      lenses.push(
+        new vscode.CodeLens(line, { title: "✓ Keep", command: "cowriting.proposalAcceptMenu", arguments: [v.id] }),
+        new vscode.CodeLens(line, { title: "✗ Reject", command: "cowriting.proposalRejectMenu", arguments: [v.id] }),
+      );
+    }
+    return lenses;
+  }
+
+  /** The dropdown: this-proposal vs all-proposals, then dispatch. */
+  private async menu(kind: "accept" | "reject", id?: string): Promise<void> {
+    const doc = vscode.window.activeTextEditor?.document;
+    if (!doc || !id) return;
+    const key = this.proposals.keyFor(doc);
+    const verb = kind === "accept" ? "Accept" : "Reject";
+    const pick = await vscode.window.showQuickPick(
+      [`${verb} this proposal`, `${verb} ALL proposals`],
+      { placeHolder: `${verb} Claude's proposal` },
+    );
+    if (!pick) return;
+    const all = pick.includes("ALL");
+    if (kind === "accept") {
+      // The batch commands own the applied/skipped reporting (#70: a silent:true
+      // batch with a discarded tally re-creates the silent-skip failure mode).
+      if (all) await vscode.commands.executeCommand("cowriting.acceptAllProposals");
+      else await this.proposals.finalizeInPlace(key, id);
+    } else {
+      if (all) await vscode.commands.executeCommand("cowriting.rejectAllProposals");
+      else await this.proposals.revertInPlace(key, id);
+    }
+  }
+
+  dispose(): void {
+    for (const t of this.pendingApply.values()) clearTimeout(t);
+    this.pendingApply.clear();
+    for (const t of this.pendingRender.values()) clearTimeout(t);
+    this.pendingRender.clear();
+    for (const d of this.disposables) d.dispose();
+  }
+}
@@ -7,7 +7,18 @@ import { ProposalController } from "./proposalController";
 import { buildFingerprint } from "./anchorer";
 import { VersionGuard } from "./versionGuard";
 import { BaselineStore } from "./baselineStore";
+import { GlobalSidecarStore } from "./globalSidecarStore";
+import { SidecarRouter } from "./sidecarRouter";
 import { DiffViewController } from "./diffViewController";
+import { GitBaselineAdapter } from "./gitBaseline";
+import { EditFlow } from "./editFlow";
+import { LiveProgressUi } from "./liveProgressUi";
+import { EditorProposalController } from "./editorProposalController";
+import { isAuthorable, routeEdit } from "./workspacePath";
+import { CoeditingRegistry } from "./coeditingRegistry";
+import { ScmSurfaceController } from "./scmSurface";
+import type MarkdownIt from "markdown-it";
+import { cowritingMarkdownItPlugin, type AnnotationInputs } from "./previewAnnotations";

 const CHANNEL_NAME = "Cowriting (Cline SDK)";

@@ -17,12 +28,30 @@ export interface CowritingApi {
  proposalController: ProposalController;
  versionGuard: VersionGuard;
  diffViewController: DiffViewController;
+  editFlow: EditFlow;
+  sidecarRouter: SidecarRouter;
+  liveProgressUi: LiveProgressUi;
+  editorProposalController: EditorProposalController;
+  coeditingRegistry: CoeditingRegistry;
+  scmSurfaceController: ScmSurfaceController;
+  /** Task 7 test seam: the REAL production `inputsFor`, so E2E exercises the
+   *  actual registry/diffView/attribution/proposals/config wiring (not a
+   *  hand-rolled reimplementation of it). */
+  previewAnnotationHost: { inputsFor(env: unknown): AnnotationInputs | undefined };
+  /** Task 7 (D3/D21): the markdown-it plugin factory VS Code's built-in preview
+   *  loads via `markdown.markdownItPlugins` (package.json). */
+  extendMarkdownIt: (md: unknown) => unknown;
 }

 export function activate(context: vscode.ExtensionContext): CowritingApi | undefined {
  // --- POC command (Feature #2), unchanged ---
  const output = vscode.window.createOutputChannel(CHANNEL_NAME);
  context.subscriptions.push(output);
+
+  // #60: shared live-progress UI (notification activity line + "Cowriting: Claude"
+  // OutputChannel) for both Ask-Claude entry points.
+  const liveProgressUi = new LiveProgressUi();
+  context.subscriptions.push(liveProgressUi);
  context.subscriptions.push(
    vscode.commands.registerCommand("cowriting.showClineSdkInfo", async () => {
      try {
@@ -44,76 +73,178 @@ export function activate(context: vscode.ExtensionContext): CowritingApi | undef
    }),
  );

-  // --- F6: diff-view toggle (Features #17 + #19) — workspace-INDEPENDENT ---
-  // F6 works on ANY diffable doc (file or untitled), so it is constructed
-  // regardless of whether a folder is open, and its commands are always live
-  // (never stubbed). Baseline lives in VS Code's per-extension GLOBAL storage
-  // (always present), never the repo (INV-19). The machine-landing advance is
-  // wired in the with-root branch below (landings only occur on workspace files
-  // through the seam). The controller self-wires baseline capture on open.
+  // --- PUC-7 (native-surfaces migration): the INV-10 opt-in gate ---
+  // Every later native surface (SCM/QuickDiff, comments, preview annotations,
+  // edit commands, decorations) checks isCoediting before attaching. Persisted
+  // in workspaceState so a reload restores the set.
+  const coeditingRegistry = new CoeditingRegistry(context.workspaceState);
+  context.subscriptions.push(coeditingRegistry);
+  context.subscriptions.push(
+    vscode.commands.registerCommand("cowriting.coeditDocument", () => {
+      const ed = vscode.window.activeTextEditor;
+      if (!ed || ed.document.languageId !== "markdown" || !isAuthorable(ed.document.uri.scheme)) {
+        void vscode.window.showWarningMessage("Cowriting: open a Markdown document to coedit it.");
+        return;
+      }
+      coeditingRegistry.enter(ed.document.uri);
+      coeditingRegistry.syncContext(ed);
+    }),
+    vscode.commands.registerCommand("cowriting.stopCoediting", () => {
+      const ed = vscode.window.activeTextEditor;
+      if (!ed) return;
+      coeditingRegistry.exit(ed.document.uri);
+      coeditingRegistry.syncContext(ed);
+    }),
+    vscode.window.onDidChangeActiveTextEditor((ed) => coeditingRegistry.syncContext(ed)),
+  );
+  coeditingRegistry.syncContext(vscode.window.activeTextEditor);
+
+  // --- Baseline router (native-surfaces migration, spec §6.4, INV-7) — workspace-
+  // INDEPENDENT --- The two-pane vscode.diff VIEW was removed in #34 (F10's
+  // rendered preview is the single review surface); only the baseline data
+  // layer survives, consumed by F7/F10. It tracks a baseline for ANY diffable
+  // doc (file or untitled), so it is constructed regardless of whether a folder
+  // is open. A SNAPSHOT baseline lives in VS Code's per-extension GLOBAL
+  // storage (always present), never the repo (INV-19); a git-tracked doc's HEAD
+  // baseline is never persisted (INV-7). Baselines are established only for
+  // documents the CoeditingRegistry has entered — see the registry wiring
+  // inside DiffViewController's constructor.
+  const gitBaseline = new GitBaselineAdapter();
+  context.subscriptions.push(gitBaseline);
  const baselineStorageDir = context.globalStorageUri?.fsPath;
  const baselineStore = baselineStorageDir ? new BaselineStore(baselineStorageDir) : null;
-  const diffViewController = new DiffViewController(baselineStore);
+  const diffViewController = new DiffViewController(baselineStore, gitBaseline, coeditingRegistry);
  context.subscriptions.push(diffViewController);

-  // --- F2: region-anchored threads (Feature #4) ---
+  // Task 1 hook: expose the resolved baseline mode ("head"/"snapshot") to the
+  // registry's context-key sync, and re-sync `cowriting.baselineMode` whenever
+  // a baseline is (re)captured (entry, head-refresh, pin) — the "Mark Changes
+  // as Reviewed" menu entry (snapshot-only, Task 3) is gated on that key.
+  coeditingRegistry.baselineModeOf = (uri) => diffViewController.modeOf(uri.toString());
+  context.subscriptions.push(
+    diffViewController.onDidChangeBaseline(() => coeditingRegistry.syncContext(vscode.window.activeTextEditor)),
+  );
+
+  // --- Task 3: the native diff surface (spec §6.4/§5, INV-13) — QuickDiff
+  // gutter bars, the cowriting-baseline: content provider, "Review Changes",
+  // and the status-bar change count. Constructed after diffViewController
+  // (consumes getBaseline/modeOf/onDidChangeBaseline) and the registry
+  // (consumes isCoediting/onDidChange), gated on INV-10. ---
+  const scmSurfaceController = new ScmSurfaceController(coeditingRegistry, diffViewController);
+  context.subscriptions.push(scmSurfaceController);
+
+  // F8: the out-of-workspace/untitled authoring sidecar — same GLOBAL storage
+  // home as the F6 baseline, keyed by sha256(uri) (INV-19/24). Constructed
+  // workspace-independently; the router falls back to it for any non-in-folder
+  // doc. When globalStorageUri is unavailable, file writes throw (authoring
+  // degrades to errors-on-write, the F6-equivalent edge) but untitled in-memory
+  // still works — PUC-5.
+  const globalSidecarStore = new GlobalSidecarStore(baselineStorageDir ?? "");
+
+  // --- F8: authoring on ANY document (in-folder, out-of-folder, untitled) ---
+  // The router routes per-document: in-workspace file: → the committable repo
+  // `.threads/` sidecar (CoauthorStore, INV-2); out-of-folder file: + untitled:
+  // → the global-storage sidecar (GlobalSidecarStore). Constructed even with NO
+  // folder open (everything then routes global) — the F6 #19 precedent, now
+  // extended to authoring (F2 threads, F3 attribution, F4 propose/accept).
  const root = vscode.workspace.workspaceFolders?.[0]?.uri.fsPath;
-  if (!root) {
-    // No folder open → nothing to anchor against, but every contributed
-    // command must still exist: leave them unregistered and the palette
-    // errors with "command not found" (#8). Register warning stubs instead;
-    // opening a folder reloads the window, re-running activate with a root.
-    const stub = () =>
-      void vscode.window.showWarningMessage(
-        "Cowriting: open a folder first — coauthoring anchors threads and attribution to workspace files.",
-      );
-    for (const command of [
-      "cowriting.createThread",
-      "cowriting.reply",
-      "cowriting.resolveThread",
-      "cowriting.reopenThread",
-      "cowriting.editSelection",
-      "cowriting.toggleAttribution",
-      "cowriting.applyAgentEdit",
-      "cowriting.acceptProposal",
-      "cowriting.rejectProposal",
-      "cowriting.proposeAgentEdit",
-    ]) {
-      context.subscriptions.push(vscode.commands.registerCommand(command, stub));
-    }
-    // F6 (toggleDiffView / pinDiffBaseline) is NOT stubbed here — it works
-    // without a folder (on untitled buffers and out-of-folder files); its real
-    // commands were registered above by DiffViewController.
-    return undefined;
-  }
-  const store = new CoauthorStore(root);
+  const coauthorStore = root ? new CoauthorStore(root) : null;
+  const sidecarRouter = new SidecarRouter(coauthorStore, globalSidecarStore, root);
  // F5 (INV-16): one shared guard — newer-major sidecars are read-only, one
  // warning per doc across the three co-owning controllers.
-  const versionGuard = new VersionGuard(store);
-  const threadController = new ThreadController(store, root, versionGuard);
-  context.subscriptions.push(threadController);
+  const versionGuard = new VersionGuard(sidecarRouter);

  // --- F3: live attribution (Feature #6) ---
-  const attributionController = new AttributionController(store, root, versionGuard);
+  const attributionController = new AttributionController(sidecarRouter, root, versionGuard, coeditingRegistry);
  context.subscriptions.push(attributionController);

  // --- F4: propose/accept (Feature #12) ---
-  const proposalController = new ProposalController(store, attributionController, root, versionGuard);
+  const proposalController = new ProposalController(
+    sidecarRouter,
+    attributionController,
+    root,
+    versionGuard,
+    coeditingRegistry,
+  );
  context.subscriptions.push(proposalController);

-  // --- F6 machine-landing wiring (with-root only) ---
-  // The seam's single machine-landing signal advances the baseline (INV-18).
-  // The seam fires only on workspace files, so this wiring belongs here; the
-  // DiffViewController itself was constructed above (workspace-independent).
+  // --- F11/F12 (native-surfaces migration, spec §6.10): the edit flow — the
+  // `cowriting.editDocument` command, the instruction-prompt/progress-wrapped
+  // "ask Claude" gesture (INV-10 gate), and the turn→proposal(s) cut
+  // (`runEditAndPropose`, INV-39/40). Constructed BEFORE the thread controller,
+  // which consumes it too (Task 6: the comment-loop's "make this edit" offer
+  // calls runEditAndPropose). ---
+  const editFlow = new EditFlow(proposalController);
+  context.subscriptions.push(editFlow);
+
+  // --- F2/F10 (Task 6, D19/D10/D8): comments-first ask + the comment→reply→
+  // offer→proposal loop. Consumes editFlow (offer → pending proposals) and
+  // liveProgressUi (the shared "asking Claude…" notification + output channel). ---
+  const threadController = new ThreadController(sidecarRouter, root, versionGuard, coeditingRegistry, editFlow, liveProgressUi);
+  context.subscriptions.push(threadController);
+
+  // --- F12 (#64): the editor surface — optimistic-apply proposals into the buffer,
+  // decorate the diff, and provide Accept ▾/Reject ▾ CodeLens (INV-48/49/52/53). ---
+  const editorProposalController = new EditorProposalController(
+    proposalController,
+    diffViewController,
+    attributionController,
+    coeditingRegistry,
+  );
+  context.subscriptions.push(editorProposalController);
+
+  // #46 (INV-42): accept every pending proposal on the active doc in one gesture
+  // (also reachable from the top-of-file "Keep all (N)" CodeLens, Task 8 PUC-2).
+  // Reuses the batched F4 seam + reports applied-vs-skipped, routing directly to
+  // ProposalController (native-surfaces migration: no preview-controller
+  // indirection).
  context.subscriptions.push(
-    attributionController.onDidApplyAgentEdit((e) => diffViewController.advance(e.document)),
+    vscode.commands.registerCommand("cowriting.acceptAllProposals", async () => {
+      const doc = vscode.window.activeTextEditor?.document;
+      if (!doc || doc.languageId !== "markdown") {
+        void vscode.window.showWarningMessage("Cowriting: open a Markdown document to accept its proposals.");
+        return;
+      }
+      const { applied, skipped } = await proposalController.acceptAllProposals(doc);
+      if (applied === 0 && skipped === 0) return;
+      const skipNote = skipped > 0 ? `, ${skipped} skipped (target text changed — undo or reject)` : "";
+      void vscode.window.showInformationMessage(
+        `Cowriting: accepted ${applied} proposal${applied === 1 ? "" : "s"}${skipNote}.`,
+      );
+    }),
  );

-  // One SHARED sidecar watcher for both controllers; self-writes are
-  // suppressed centrally in the store (the sidecar is co-owned).
+  // #64 (INV-53): reject every pending proposal on the active doc in one gesture
+  // (native-surfaces migration: routes directly to ProposalController).
+  context.subscriptions.push(
+    vscode.commands.registerCommand("cowriting.rejectAllProposals", async () => {
+      const doc = vscode.window.activeTextEditor?.document;
+      if (!doc || doc.languageId !== "markdown") {
+        void vscode.window.showWarningMessage("Cowriting: open a Markdown document to reject its proposals.");
+        return;
+      }
+      const { reverted, skipped } = await proposalController.rejectAll(doc);
+      if (reverted === 0 && skipped === 0) return;
+      const skipNote = skipped > 0 ? `, ${skipped} skipped (applied text not locatable — undo your edits or Keep)` : "";
+      void vscode.window.showInformationMessage(
+        `Cowriting: rejected ${reverted} proposal${reverted === 1 ? "" : "s"}${skipNote}.`,
+      );
+    }),
+  );
+
+  // NOTE (native-surfaces migration, INV-7/INV-18 retirement): the shipped
+  // machine-landing baseline advance (#48) — wiring
+  // attributionController.onDidApplyAgentEdit → diffViewController.advance —
+  // was removed here. A landed Claude edit now stays a visible change-since-
+  // baseline until the file is committed (head mode) or reviewed via
+  // "Mark Changes as Reviewed" (snapshot mode).
+
+  // One SHARED sidecar watcher for both controllers; self-writes are suppressed
+  // centrally in the repo store (only repo `.threads/` sidecars are watched —
+  // global artifacts live outside the workspace). Harmless when no folder is open.
  const watcher = vscode.workspace.createFileSystemWatcher("**/.threads/**/*.json");
  const onSidecar = (uri: vscode.Uri) => {
-    if (store.consumeSelfWrite(uri.fsPath)) return;
+    if (sidecarRouter.consumeSelfWrite(uri.fsPath)) return;
    threadController.handleExternalSidecarChange(uri);
    attributionController.handleExternalSidecarChange(uri);
    proposalController.handleExternalSidecarChange(uri);
@@ -169,88 +300,116 @@ export function activate(context: vscode.ExtensionContext): CowritingApi | undef
    ),
  );

-  // F3 SLICE-5 → F4 SLICE-4: the live turn — selection + instruction → one
-  // claude-code SDK turn (liveTurn.ts, INV-8) → a PENDING PROPOSAL (F4,
-  // INV-10); the applyAgentEdit seam (INV-9) now fires only on accept.
+  // Task 6 (D19): the comments-first ask replaces the old inline turn
+  // plumbing here — the ask now IS a comment. threadController.askClaude()
+  // opens a focused comment box on the current selection (or, with no
+  // selection, a top-anchored whole-document thread); the comment→reply→
+  // offer→proposal loop (ThreadController.respondInThread/makeThreadEdit)
+  // takes it from there. Kept as its own command (rather than folded into
+  // cowriting.edit) for the host E2E harness and the internal seams.
  context.subscriptions.push(
    vscode.commands.registerCommand("cowriting.editSelection", async () => {
+      await threadController.askClaude();
+    }),
+  );
+
+  // Code-review fix (Finding 1, session native-surfaces-exec): `cowriting.editDocument`
+  // used to delegate to `EditFlow.askClaude(doc, {kind:"document"})`, which prompted via
+  // `askEditInstruction` — a rejecting stub after Task 6 sunset the input webview
+  // (spec §6.10; both `askClaude`/`askEditInstruction` were deleted from EditFlow in
+  // Task 8 once their only caller, the review webview, died too). Reached with no
+  // selection (the common case for `cowriting.edit`), that was a SILENT dead end: the
+  // reject fired before EditFlow.askClaude's own try/catch and was `void`'d, so no
+  // toast, no thread, nothing. This command now resolves
+  // its target document exactly as before (a tab's clicked URI, falling back to the
+  // active editor), focuses it, and hands off to the SAME comments-first ask as
+  // `cowriting.editSelection` — `ThreadController.askClaude()` top-anchors a
+  // whole-document ask when the (now-focused) editor's selection is empty (D19). Kept
+  // registered + hidden from the palette for the host E2E harness and #42's
+  // tab-targeting behavior; `EditFlow.runEditAndPropose` (untouched) stays the E2E seam
+  // for driving a turn without the native comment UI — see f12Reach.test.ts /
+  // f11Toolbar.test.ts / f12Accept.test.ts / f12Review.test.ts.
+  context.subscriptions.push(
+    vscode.commands.registerCommand("cowriting.editDocument", async (uri?: vscode.Uri) => {
+      const doc = uri
+        ? vscode.workspace.textDocuments.find((d) => d.uri.toString() === uri.toString()) ??
+          (await vscode.workspace.openTextDocument(uri))
+        : vscode.window.activeTextEditor?.document;
+      if (!doc || doc.languageId !== "markdown") {
+        void vscode.window.showWarningMessage("Cowriting: open a Markdown document to ask Claude to edit it.");
+        return;
+      }
+      if (vscode.window.activeTextEditor?.document.uri.toString() !== doc.uri.toString()) {
+        await vscode.window.showTextDocument(doc);
+      }
+      await threadController.askClaude();
+    }),
+  );
+
+  // The single user-facing "Ask Claude to Edit" gesture (one command, one
+  // keybinding, one menu entry). It routes to the selection flow (editSelection)
+  // or the whole-document flow (editDocument) by selection/context — see
+  // routeEdit. Both underlying commands now end at ThreadController.askClaude()
+  // (comments-first ask, D19); they stay registered for the host E2E harness and
+  // the internal seams, but are hidden from the palette.
+  context.subscriptions.push(
+    vscode.commands.registerCommand("cowriting.edit", async (uri?: vscode.Uri) => {
      const editor = vscode.window.activeTextEditor;
-      if (
-        !editor ||
-        editor.selection.isEmpty ||
-        editor.document.uri.scheme !== "file" ||
-        !editor.document.uri.fsPath.startsWith(root)
-      ) {
-        void vscode.window.showWarningMessage("Cowriting: select some text in a workspace document first.");
-        return;
-      }
-      const instruction = await vscode.window.showInputBox({
-        prompt: "What should Claude do with the selection?",
-        placeHolder: "e.g. tighten this paragraph",
+      const route = routeEdit({
+        hasUri: !!uri,
+        uriMatchesActiveEditor: !!uri && editor?.document.uri.toString() === uri.toString(),
+        hasActiveEditor: !!editor,
+        selectionEmpty: editor?.selection.isEmpty ?? true,
      });
-      if (!instruction) return;
-      if (editor.selection.isEmpty) {
-        void vscode.window.showWarningMessage("Cowriting: select some text in a workspace document first.");
-        return;
-      }
-      const document = editor.document;
-      const selection = editor.selection;
-      const selectedText = document.getText(selection);
-      // Capture the anchor BEFORE the turn (spec §6.5 PUC-1): mid-turn edits
-      // can't skew it — the proposal renders wherever the target re-resolves.
-      const fp = buildFingerprint(document.getText(), {
-        start: document.offsetAt(selection.start),
-        end: document.offsetAt(selection.end),
-      });
-      const turnId = `turn-${Date.now().toString(36)}`;
-      try {
-        await vscode.window.withProgress(
-          { location: vscode.ProgressLocation.Notification, title: "Cowriting: asking Claude…" },
-          async () => {
-            const { runEditTurn } = await import("./liveTurn");
-            const turn = await runEditTurn(instruction, selectedText);
-            if (turn.replacement === "") {
-              void vscode.window.showWarningMessage(
-                "Cowriting: Claude returned an empty replacement — nothing was proposed.",
-              );
-              return;
-            }
-            if (turn.replacement === selectedText) {
-              void vscode.window.showInformationMessage(
-                "Cowriting: Claude proposed no change to the selection.",
-              );
-              return;
-            }
-            // F4 (INV-10): the turn ends in a PROPOSAL, not a buffer mutation.
-            // The seam now fires only on accept (ProposalController, INV-9).
-            const id = await proposalController.propose(
-              document,
-              fp,
-              turn.replacement,
-              {
-                kind: "agent",
-                id: "claude",
-                agent: { sdk: "@cline/sdk", model: turn.model, sessionId: turn.sessionId },
-              },
-              { turnId, instruction },
-            );
-            if (id) {
-              void vscode.window.showInformationMessage(
-                "Cowriting: Claude proposed an edit — review the diff at the highlighted range (✓ accept / ✗ reject).",
-              );
-            }
-          },
-        );
-      } catch (err) {
-        const message = err instanceof Error ? err.message : String(err);
-        void vscode.window.showErrorMessage(`Cowriting: Claude edit failed — ${message}`);
+      if (route === "selection") {
+        await vscode.commands.executeCommand("cowriting.editSelection");
+      } else {
+        await vscode.commands.executeCommand("cowriting.editDocument", uri);
      }
    }),
  );

+  // Task 8 (native-surfaces migration, spec §6.10): the minimal gateway replacing
+  // the sunset review-webview's #41 right-click entries. Resolves the clicked
+  // doc (or falls back to the active editor, mirroring editDocument/#41), enters
+  // coediting if not already (so a fresh doc gets a baseline to review against —
+  // no-op if already entered or not authorable), then hands off to VS Code's
+  // OWN "Open Preview to the Side" — the built-in preview IS the review surface
+  // now (Task 7's annotations render inside it).
+  context.subscriptions.push(
+    vscode.commands.registerCommand("cowriting.openReviewPreview", async (uri?: vscode.Uri) => {
+      const doc = uri
+        ? vscode.workspace.textDocuments.find((d) => d.uri.toString() === uri.toString()) ??
+          (await vscode.workspace.openTextDocument(uri))
+        : vscode.window.activeTextEditor?.document;
+      if (!doc || doc.languageId !== "markdown") {
+        void vscode.window.showWarningMessage("Cowriting: open a Markdown document to review it.");
+        return;
+      }
+      if (vscode.window.activeTextEditor?.document.uri.toString() !== doc.uri.toString()) {
+        await vscode.window.showTextDocument(doc);
+      }
+      if (!coeditingRegistry.isCoediting(doc.uri) && isAuthorable(doc.uri.scheme)) {
+        await vscode.commands.executeCommand("cowriting.coeditDocument");
+      }
+      await vscode.commands.executeCommand("markdown.showPreviewToSide", doc.uri);
+    }),
+  );
+
  // Render threads + attributions for already-open editors, and on future opens.
+  // INV-10: gated on the registry too — belt-and-suspenders with the per-controller
+  // gates (renderAll/loadAll already early-return), but keeps the intent explicit
+  // at the call site and avoids the no-op calls for a non-entered doc.
  const renderIfOpen = (doc: vscode.TextDocument) => {
-    if (doc.uri.scheme === "file" && doc.uri.fsPath.startsWith(root)) {
+    if (isAuthorable(doc.uri.scheme) && coeditingRegistry.isCoediting(doc.uri)) {
+      // Finding 3 (final whole-branch review): DiffViewController only
+      // establishes a baseline on the registry's enter TRANSITION and, at
+      // startup, for documents already open. A doc that is a PERSISTED
+      // coediting member but wasn't open yet at either of those moments (e.g.
+      // lazily restored after a window reload) reaches this handler with no
+      // baseline/mode — establish it now so QuickDiff/Review-Changes/Mark-
+      // Reviewed never see a silent empty-baseline gap.
+      if (diffViewController.modeOf(doc.uri.toString()) === undefined) void diffViewController.establish(doc);
      threadController.renderAll(doc);
      attributionController.loadAll(doc);
      proposalController.renderAll(doc);
@@ -259,7 +418,79 @@ export function activate(context: vscode.ExtensionContext): CowritingApi | undef
  vscode.workspace.textDocuments.forEach(renderIfOpen);
  context.subscriptions.push(vscode.workspace.onDidOpenTextDocument(renderIfOpen));

-  return { threadController, attributionController, proposalController, versionGuard, diffViewController };
+  // --- Task 7 (D3/D21, PUC-3): annotations in the BUILT-IN Markdown preview ---
+  // `env.currentDocument` (VS Code's markdown-it render env) is an OBSERVED
+  // preview-engine behavior, not a documented contract, so `lastCoeditedUri`
+  // is the fallback that keeps single-doc correctness when it's absent — the
+  // last editor that WAS coediting when it lost focus (e.g. focus moved to the
+  // preview pane itself) stays the annotation target until a different
+  // coedited doc becomes active.
+  let lastCoeditedUri: string | undefined;
+  const trackLastCoedited = (ed: vscode.TextEditor | undefined) => {
+    if (ed && coeditingRegistry.isCoediting(ed.document.uri)) lastCoeditedUri = ed.document.uri.toString();
+  };
+  trackLastCoedited(vscode.window.activeTextEditor);
+  context.subscriptions.push(
+    vscode.window.onDidChangeActiveTextEditor(trackLastCoedited),
+    // The active editor doesn't itself change when `cowriting.coeditDocument`
+    // enters IT into coediting — re-check on every gate change too, so the
+    // very first render after entering already has an annotation target.
+    coeditingRegistry.onDidChange(() => trackLastCoedited(vscode.window.activeTextEditor)),
+  );
+
+  const previewAnnotationHost = {
+    inputsFor(env: unknown): AnnotationInputs | undefined {
+      const envUri = (env as { currentDocument?: { toString(): string } } | undefined)?.currentDocument?.toString();
+      // Finding 1 (final whole-branch review): the `lastCoeditedUri` fallback
+      // is for an ABSENT env only (env.currentDocument didn't resolve) — a
+      // present envUri is authoritative for the doc actually being previewed
+      // and must never fall back to a different (or since-exited) document's
+      // inputs. Either way, the winning key must currently be a coediting
+      // member (INV-10) — an exited doc's stale `lastCoeditedUri` fails this
+      // check too, closing the "preview stays annotated after stopCoediting"
+      // gap.
+      const key = envUri === undefined ? lastCoeditedUri : envUri;
+      if (!key || !coeditingRegistry.list().includes(key)) return undefined;
+      const doc = vscode.workspace.textDocuments.find((d) => d.uri.toString() === key);
+      if (!doc) return undefined;
+      const baseline = diffViewController.getBaseline(key);
+      return {
+        baselineText: baseline?.text,
+        baselineReason: baseline?.reason,
+        spans: attributionController.spansFor(doc),
+        proposals: proposalController.listProposals(doc),
+        enabled: vscode.workspace.getConfiguration("cowriting").get<boolean>("annotations", true),
+      };
+    },
+  };
+
+  // Toggle: flips `cowriting.annotations` + refreshes the built-in preview
+  // (VS Code re-invokes extendMarkdownIt's plugin on the next render — no
+  // extension-side re-render call exists beyond the standard refresh command).
+  context.subscriptions.push(
+    vscode.commands.registerCommand("cowriting.toggleAnnotations", async () => {
+      const config = vscode.workspace.getConfiguration("cowriting");
+      const current = config.get<boolean>("annotations", true);
+      await config.update("annotations", !current, vscode.ConfigurationTarget.Global);
+      await vscode.commands.executeCommand("markdown.preview.refresh");
+    }),
+  );
+
+  return {
+    threadController,
+    attributionController,
+    proposalController,
+    versionGuard,
+    diffViewController,
+    editFlow,
+    sidecarRouter,
+    liveProgressUi,
+    editorProposalController,
+    coeditingRegistry,
+    scmSurfaceController,
+    previewAnnotationHost,
+    extendMarkdownIt: (md: unknown) => cowritingMarkdownItPlugin(md as MarkdownIt, previewAnnotationHost),
+  };
 }

 export function deactivate(): void {
@@ -0,0 +1,136 @@
+/**
+ * GitBaselineAdapter — resolves a document's git-HEAD baseline via the built-in
+ * Git extension API (spec §6.4, INV-7). Spike-verified: Repository.show('HEAD',
+ * path) + state.onDidChange work as assumed; the change event is coarse, so
+ * re-reads key off state.HEAD.commit; a nested repo may need openRepository().
+ *
+ * Hardening (Task 2 E2E, host-verified): for a repo that is NOT a workspace
+ * folder (the common case here — F8/#19 precedent, coediting works on any
+ * file), the git extension's own filesystem watcher does not reliably notice
+ * an out-of-band commit (e.g. a `git commit` run from a terminal, or in the
+ * host E2E) — `repo.state.onDidChange` can simply never fire. We additionally
+ * watch `.git/logs/HEAD` (the reflog VS Code itself also watches for
+ * in-workspace repos) with Node's own `fs.watch` and nudge `repo.status()`,
+ * which reliably re-syncs `state.HEAD` and lets the existing onDidChange path
+ * do the rest. Best-effort: any failure here just falls back to the passive
+ * listener.
+ */
+import * as fs from "node:fs";
+import * as path from "node:path";
+import * as vscode from "vscode";
+
+interface GitRepository {
+  rootUri: vscode.Uri;
+  state: { HEAD?: { commit?: string }; onDidChange: vscode.Event<void> };
+  show(ref: string, path: string): Promise<string>;
+  status(): Promise<void>;
+}
+interface GitAPI {
+  repositories: GitRepository[];
+  getRepository(uri: vscode.Uri): GitRepository | null;
+  openRepository(root: vscode.Uri): Promise<GitRepository | null>;
+  onDidOpenRepository: vscode.Event<GitRepository>;
+}
+
+export interface HeadBaseline { text: string; commit: string }
+
+export class GitBaselineAdapter implements vscode.Disposable {
+  private api: GitAPI | null | undefined;      // undefined = not resolved yet; null = unavailable
+  private readonly hooked = new WeakSet<GitRepository>();
+  private readonly lastCommit = new Map<string, string>();   // repo root → HEAD commit
+  private readonly emitter = new vscode.EventEmitter<void>();
+  readonly onDidChangeHead = this.emitter.event;
+  private readonly disposables: vscode.Disposable[] = [this.emitter];
+
+  private async gitApi(): Promise<GitAPI | null> {
+    if (this.api !== undefined) return this.api;
+    try {
+      const ext = vscode.extensions.getExtension("vscode.git");
+      if (!ext) return (this.api = null);
+      const exports = ext.isActive ? ext.exports : await ext.activate();
+      this.api = exports.getAPI(1) as GitAPI;
+      this.disposables.push(this.api.onDidOpenRepository((r) => this.hook(r)));
+      for (const r of this.api.repositories) this.hook(r);
+    } catch {
+      this.api = null;
+    }
+    return this.api;
+  }
+
+  /** HEAD blob text + commit for a tracked file; null when untracked / no repo / no HEAD. */
+  async headFor(uri: vscode.Uri): Promise<HeadBaseline | null> {
+    if (uri.scheme !== "file") return null;
+    const git = await this.gitApi();
+    if (!git) return null;
+    let repo = git.getRepository(uri);
+    if (!repo) repo = await this.discoverNested(git, uri);
+    if (!repo) return null;
+    this.hook(repo);
+    try {
+      const text = await repo.show("HEAD", uri.fsPath);
+      return { text, commit: repo.state.HEAD?.commit ?? "" };
+    } catch {
+      return null;    // untracked, ignored, or no commit yet → snapshot mode
+    }
+  }
+
+  /** Spike finding: a nested repo may not be auto-opened — walk up for .git, then openRepository. */
+  private async discoverNested(git: GitAPI, uri: vscode.Uri): Promise<GitRepository | null> {
+    let dir = path.dirname(uri.fsPath);
+    for (let i = 0; i < 32; i++) {
+      if (fs.existsSync(path.join(dir, ".git"))) {
+        await git.openRepository(vscode.Uri.file(dir));
+        return git.getRepository(uri);
+      }
+      const parent = path.dirname(dir);
+      if (parent === dir) break;
+      dir = parent;
+    }
+    return null;
+  }
+
+  /** Watch a repo; fire onDidChangeHead only when HEAD's commit actually moves. */
+  private hook(repo: GitRepository): void {
+    if (this.hooked.has(repo)) return;
+    this.hooked.add(repo);
+    this.lastCommit.set(repo.rootUri.toString(), repo.state.HEAD?.commit ?? "");
+    this.disposables.push(
+      repo.state.onDidChange(() => {
+        const key = repo.rootUri.toString();
+        const commit = repo.state.HEAD?.commit ?? "";
+        if (this.lastCommit.get(key) !== commit) {
+          this.lastCommit.set(key, commit);
+          this.emitter.fire();
+        }
+      }),
+    );
+    this.watchReflog(repo);
+  }
+
+  /** Best-effort nudge: watch `.git/logs/HEAD` directly and force `repo.status()`
+   *  on any write, so an out-of-band commit is picked up even when the git
+   *  extension's own watcher doesn't fire (see the class doc). */
+  private watchReflog(repo: GitRepository): void {
+    try {
+      const reflog = path.join(repo.rootUri.fsPath, ".git", "logs", "HEAD");
+      if (!fs.existsSync(reflog)) return;
+      let pending: ReturnType<typeof setTimeout> | undefined;
+      const watcher = fs.watch(reflog, () => {
+        if (pending) clearTimeout(pending);
+        pending = setTimeout(() => void repo.status().catch(() => {}), 150);
+      });
+      this.disposables.push({
+        dispose: () => {
+          watcher.close();
+          if (pending) clearTimeout(pending);
+        },
+      });
+    } catch {
+      // best-effort only — the passive repo.state.onDidChange listener still applies.
+    }
+  }
+
+  dispose(): void {
+    for (const d of this.disposables) d.dispose();
+  }
+}
@@ -0,0 +1,89 @@
+/**
+ * GlobalSidecarStore — the out-of-workspace/untitled authoring sidecar (F8 spec
+ * §6.2/§6.4), mirroring BaselineStore (src/baselineStore.ts): vscode-free (Node
+ * fs/crypto only), one `Artifact` JSON per document in VS Code's per-extension
+ * GLOBAL storage, NEVER the repo (INV-19/24). The key passed in is the DOCUMENT
+ * KEY (the URI string for the docs this store serves):
+ *   - `file:` URI  → disk at `<dir>/sidecars/<sha256(uri)>.json`.
+ *   - `untitled:`  → an in-memory Map only (no durable identity — the F6
+ *     degrade); lost on reload, and never read by mergeArtifacts (INV-25).
+ * `consumeSelfWrite` is a no-op (these artifacts are outside the `.threads/`
+ * watcher, so there is no self-write storm to suppress).
+ */
+import * as fs from "node:fs";
+import * as path from "node:path";
+import { createHash } from "node:crypto";
+import {
+  SCHEMA_VERSION,
+  emptyArtifact,
+  isNewerMajor,
+  serializeArtifact,
+  type Artifact,
+} from "./model";
+import type { SidecarStore } from "./sidecarStore";
+
+export class GlobalSidecarStore implements SidecarStore {
+  /** untitled keys live here only — no durable identity to persist against. */
+  private readonly memory = new Map<string, Artifact>();
+
+  /** @param storageDir absolute VS Code GLOBAL-storage dir (context.globalStorageUri.fsPath). */
+  constructor(private readonly storageDir: string) {}
+
+  private isUntitled(key: string): boolean {
+    return key.startsWith("untitled:");
+  }
+
+  /** Filesystem-safe storage key for a persistable (file:) doc: sha256 of its URI. */
+  private storageKey(key: string): string {
+    return createHash("sha256").update(key).digest("hex");
+  }
+
+  /** Disk path for a `file:` key, or undefined for an in-memory untitled key. */
+  sidecarPath(key: string): string | undefined {
+    if (this.isUntitled(key)) return undefined;
+    return path.join(this.storageDir, "sidecars", `${this.storageKey(key)}.json`);
+  }
+
+  load(key: string): Artifact | null {
+    if (this.isUntitled(key)) return this.memory.get(key) ?? null;
+    const p = this.sidecarPath(key)!;
+    if (!fs.existsSync(p)) return null;
+    return JSON.parse(fs.readFileSync(p, "utf8")) as Artifact;
+  }
+
+  save(key: string, artifact: Artifact): void {
+    if (this.isUntitled(key)) {
+      this.memory.set(key, artifact);
+      return;
+    }
+    const p = this.sidecarPath(key)!;
+    fs.mkdirSync(path.dirname(p), { recursive: true });
+    fs.writeFileSync(p, serializeArtifact(artifact), "utf8");
+  }
+
+  /** See CoauthorStore.update — same INV-16 throw + anchor prune, no self-write mark. */
+  update(key: string, mutate: (artifact: Artifact) => void): Artifact {
+    const artifact = this.load(key) ?? emptyArtifact(key);
+    if (isNewerMajor(artifact)) {
+      throw new Error(
+        `refusing to write ${key}: sidecar schemaVersion ${artifact.schemaVersion} > supported ${SCHEMA_VERSION} (INV-16)`,
+      );
+    }
+    mutate(artifact);
+    const referenced = new Set<string>([
+      ...artifact.threads.map((t) => t.anchorId),
+      ...artifact.attributions.map((a) => a.anchorId),
+      ...artifact.proposals.map((p) => p.anchorId),
+    ]);
+    for (const id of Object.keys(artifact.anchors)) {
+      if (!referenced.has(id)) delete artifact.anchors[id];
+    }
+    this.save(key, artifact);
+    return artifact;
+  }
+
+  /** No-op: global artifacts are outside the `.threads/` watcher. */
+  consumeSelfWrite(_fsPath: string): boolean {
+    return false;
+  }
+}
@@ -0,0 +1,65 @@
+/**
+ * liveProgressUi.ts — host-side relay of TurnProgress snapshots to VS Code UI
+ * (#60, spec coauthoring-live-progress.md §3.4). The ONLY surfaces are the
+ * existing withProgress notification (an activity line) and a dedicated
+ * OutputChannel streaming the full assistant text — no new network/webview
+ * surface (INV-45). vscode-only; all pure logic lives in turnProgress.ts.
+ */
+import * as vscode from "vscode";
+import { formatProgressLine, type TurnProgressSnapshot } from "./turnProgress";
+
+const CHANNEL_NAME = "Cowriting: Claude";
+
+export interface TurnUi {
+  /** Pass to runEditTurn's opts.onProgress. */
+  onProgress: (snapshot: TurnProgressSnapshot) => void;
+  /** Pass to runEditTurn's opts.signal — fired when the user cancels the notification. */
+  signal: AbortSignal;
+}
+
+export class LiveProgressUi {
+  readonly channel: vscode.OutputChannel;
+
+  constructor() {
+    this.channel = vscode.window.createOutputChannel(CHANNEL_NAME);
+  }
+
+  /**
+   * Begin one turn's UI. Writes the per-turn header to the channel and returns
+   * the onProgress relay + an AbortSignal linked to the notification's cancel
+   * token. The channel APPENDS (it doubles as a debug log of recent turns, spec
+   * §3.5); it auto-reveals (without stealing focus) on the first streamed text,
+   * gated by `cowriting.liveProgress.revealOutput`.
+   */
+  begin(
+    instruction: string,
+    progress: vscode.Progress<{ message?: string }>,
+    token: vscode.CancellationToken,
+  ): TurnUi {
+    const controller = new AbortController();
+    token.onCancellationRequested(() => controller.abort());
+    this.channel.appendLine(`── asking: ${instruction} ──`);
+
+    let revealed = false;
+    const reveal = (): void => {
+      if (revealed) return;
+      revealed = true;
+      const cfg = vscode.workspace.getConfiguration("cowriting");
+      if (cfg.get<boolean>("liveProgress.revealOutput", true)) this.channel.show(true);
+    };
+
+    const onProgress = (s: TurnProgressSnapshot): void => {
+      progress.report({ message: formatProgressLine(s) });
+      if (s.textDelta) {
+        this.channel.append(s.textDelta);
+        reveal();
+      }
+    };
+
+    return { onProgress, signal: controller.signal };
+  }
+
+  dispose(): void {
+    this.channel.dispose();
+  }
+}
@@ -9,6 +9,9 @@
 * never bundled (esbuild keeps it external).
 */

+import type { TurnProgressSnapshot } from "./turnProgress";
+import { createTurnProgressState, reduceTurnProgress } from "./turnProgress";
+
 export interface EditTurnResult {
  replacement: string;
  model: string;
@@ -16,6 +19,19 @@ export interface EditTurnResult {
  sessionId: string;
 }

+/**
+ * Options for runEditTurn. Both new fields are purely additive observability /
+ * control over the existing turn (INV-44): `onProgress` streams reduced progress
+ * snapshots out; `signal` cancels the turn in. Neither touches the result path,
+ * and neither pulls `vscode` into this module (INV-43 — `AbortSignal` is a web
+ * standard, the progress snapshot is a domain type).
+ */
+export interface RunEditTurnOptions {
+  modelId?: string;
+  onProgress?: (snapshot: TurnProgressSnapshot) => void;
+  signal?: AbortSignal;
+}
+
 const SYSTEM_PROMPT = [
  "You are a precise text editor embedded in VS Code.",
  "You will be given a piece of text and an instruction.",
@@ -40,7 +56,7 @@ export function extractReplacement(outputText: string, selectedText: string): st
 export async function runEditTurn(
  instruction: string,
  selectedText: string,
-  opts?: { modelId?: string },
+  opts?: RunEditTurnOptions,
 ): Promise<EditTurnResult> {
  const sdk = await import("@cline/sdk");
  const modelId = opts?.modelId ?? "sonnet";
@@ -48,17 +64,70 @@ export async function runEditTurn(
    providerId: "claude-code",
    modelId,
    systemPrompt: SYSTEM_PROMPT,
+    // The claude-code provider spawns the bundled `claude` CLI. On macOS, when
+    // that process is launched from inside VS Code it performs IDE auto-connect,
+    // which shells out to `osascript` (`tell application …`) to identify/activate
+    // the editor — raising the macOS "control other applications" (Apple Events)
+    // permission prompt (#59). We only consume the turn's text result and never
+    // use the IDE connection, so we disable it: `CLAUDE_CODE_AUTO_CONNECT_IDE=0`
+    // hard-returns out of the auto-connect path in the binary, and
+    // `CLAUDE_CODE_IDE_SKIP_AUTO_INSTALL=1` skips the extension-install deep link.
+    // Seeding with the full `process.env` keeps the default inheritance intact
+    // (login under HOME, proxy/CA vars) — we only add the two switches. The
+    // provider forwards `options.env` into the spawned child (verified end to end
+    // through @cline/sdk → ai-sdk-provider-claude-code → @anthropic-ai/claude-agent-sdk).
+    options: {
+      env: {
+        ...process.env,
+        CLAUDE_CODE_AUTO_CONNECT_IDE: "0",
+        CLAUDE_CODE_IDE_SKIP_AUTO_INSTALL: "1",
+      },
+    },
  });
-  const result = await agent.run(
-    `<instruction>\n${instruction}\n</instruction>\n<text>\n${selectedText}\n</text>`,
-  );
-  // The SDK's AgentRunResult.status union is "completed" | "aborted" | "failed"
-  // (@cline/shared agent.d.ts) — "completed" is the success status.
-  if (result.status !== "completed") {
-    throw new Error(
-      `claude-code turn ${result.status}: ${result.error?.message ?? "unknown error"} ` +
-        "(is Claude Code installed and signed in?)",
+
+  // Stream reduced progress snapshots out (INV-44 additive) and wire cancellation
+  // in via the AbortSignal (INV-47). agent.subscribe returns its unsubscribe fn.
+  let state = createTurnProgressState();
+  const unsubscribe = opts?.onProgress
+    ? agent.subscribe((event) => {
+        const next = reduceTurnProgress(state, event);
+        state = next.state;
+        // Observability must never affect the result (INV-44): a throwing relay
+        // is swallowed, not allowed to propagate into the SDK and fail the turn.
+        if (next.snapshot) {
+          try {
+            opts.onProgress!(next.snapshot);
+          } catch {
+            /* progress is best-effort */
+          }
+        }
+      })
+    : undefined;
+  const onAbort = () => agent.abort();
+  opts?.signal?.addEventListener("abort", onAbort);
+
+  try {
+    // A signal already aborted before the turn starts can't be honored by
+    // agent.abort() (the SDK's AbortController isn't created until run()), so
+    // short-circuit to the same aborted outcome the call site reflects (INV-47).
+    if (opts?.signal?.aborted) {
+      throw new Error("claude-code turn aborted: cancelled before start");
+    }
+    const result = await agent.run(
+      `<instruction>\n${instruction}\n</instruction>\n<text>\n${selectedText}\n</text>`,
    );
+    // The SDK's AgentRunResult.status union is "completed" | "aborted" | "failed"
+    // (@cline/shared agent.d.ts) — "completed" is the success status. An aborted
+    // turn (user cancel) falls into this throw; the call site reflects "cancelled".
+    if (result.status !== "completed") {
+      throw new Error(
+        `claude-code turn ${result.status}: ${result.error?.message ?? "unknown error"} ` +
+          "(is Claude Code installed and signed in?)",
+      );
+    }
+    return { replacement: extractReplacement(result.outputText, selectedText), model: modelId, sessionId: result.runId };
+  } finally {
+    unsubscribe?.();
+    opts?.signal?.removeEventListener("abort", onAbort);
  }
-  return { replacement: extractReplacement(result.outputText, selectedText), model: modelId, sessionId: result.runId };
 }
@@ -0,0 +1,49 @@
+/**
+ * mermaidDiff — F7.1 (#22) dispatcher. Detects a mermaid diagram's type and
+ * routes a baseline→current diff to the matching pure differ, which re-emits the
+ * CURRENT diagram source augmented with mermaid styling directives (INV-29). Any
+ * unsupported type or parser surprise degrades to the v1 whole-block badge
+ * (INV-30) — this function never throws. vscode-free, DOM-free, deterministic.
+ */
+import { diffFlowchart } from "./mermaidFlowchartDiff";
+import { diffSequence } from "./mermaidSequenceDiff";
+
+/** Theme-neutral colors baked into emitted mermaid source (source can't read CSS vars). */
+export const CW_COLORS = {
+  added: "#2ea043",
+  changed: "#d29922",
+  removed: "#808080",
+} as const;
+
+export type DiagramType = "flowchart" | "sequence" | "other";
+
+export interface MermaidDiffAugmented {
+  kind: "augmented";
+  source: string;
+}
+export interface MermaidDiffFallback {
+  kind: "fallback";
+}
+export type MermaidDiffResult = MermaidDiffAugmented | MermaidDiffFallback;
+
+export function detectDiagramType(source: string): DiagramType {
+  for (const line of source.split(/\r?\n/)) {
+    const t = line.trim();
+    if (t === "" || t.startsWith("%%")) continue;
+    if (/^(flowchart|graph)\b/.test(t)) return "flowchart";
+    if (/^sequenceDiagram\b/.test(t)) return "sequence";
+    return "other";
+  }
+  return "other";
+}
+
+export function diffMermaid(beforeSrc: string, currentSrc: string): MermaidDiffResult {
+  const type = detectDiagramType(currentSrc);
+  try {
+    if (type === "flowchart") return { kind: "augmented", source: diffFlowchart(beforeSrc, currentSrc) };
+    if (type === "sequence") return { kind: "augmented", source: diffSequence(beforeSrc, currentSrc) };
+  } catch {
+    return { kind: "fallback" };
+  }
+  return { kind: "fallback" };
+}
@@ -0,0 +1,154 @@
+/**
+ * mermaidFlowchartDiff — F7.1 (#22). Pure flowchart parser + diff/emit. Parses a
+ * `graph`/`flowchart` source into nodes (by id) and edges (by declaration order),
+ * diffs baseline vs current, and re-emits the CURRENT source augmented with
+ * `classDef`/`class`/`linkStyle` directives coloring added/changed elements and
+ * ghosting removed ones (INV-29/31). Deterministic; no vscode, no DOM.
+ */
+import { CW_COLORS } from "./mermaidDiff";
+
+export interface FlowNode {
+  id: string;
+  label?: string;
+  open?: string;
+  close?: string;
+  /** Verbatim declaration token, e.g. `A[Start]`, for ghost re-injection. */
+  decl?: string;
+}
+export interface FlowEdge {
+  from: string;
+  to: string;
+  label?: string;
+  index: number;
+  /** Verbatim edge statement for ghost re-injection. */
+  raw: string;
+}
+export interface FlowGraph {
+  header: string;
+  nodes: Map<string, FlowNode>;
+  edges: FlowEdge[];
+}
+
+const NODE_TOKEN = /^([A-Za-z0-9_]+)(\[\[[^\]]*\]\]|\[[^\]]*\]|\(\([^)]*\)\)|\([^)]*\)|\{[^}]*\}|>[^\]]*\])?/;
+// link operators: -->, ---, -.->, -.-, ==>, ===, --x, --o, ==x, ==o, optionally |label|
+const LINK = /^(-->|---|-\.->|-\.-|==>|===|--[xo]|==[xo])(\|([^|]*)\|)?/;
+
+function shapeOf(bracket: string): { label: string; open: string; close: string } {
+  if (bracket.startsWith("[[")) return { label: bracket.slice(2, -2), open: "[[", close: "]]" };
+  if (bracket.startsWith("((")) return { label: bracket.slice(2, -2), open: "((", close: "))" };
+  const open = bracket[0];
+  const close = bracket[bracket.length - 1];
+  return { label: bracket.slice(1, -1), open, close };
+}
+
+function recordNode(nodes: Map<string, FlowNode>, token: string): string {
+  const m = token.match(NODE_TOKEN)!;
+  const id = m[1];
+  const existing = nodes.get(id) ?? { id };
+  if (m[2]) {
+    const s = shapeOf(m[2]);
+    existing.label = s.label;
+    existing.open = s.open;
+    existing.close = s.close;
+    existing.decl = `${id}${m[2]}`;
+  }
+  nodes.set(id, existing);
+  return id;
+}
+
+export function parseFlowchart(source: string): FlowGraph {
+  const lines = source.split(/\r?\n/);
+  let header = "";
+  const nodes = new Map<string, FlowNode>();
+  const edges: FlowEdge[] = [];
+  for (const rawLine of lines) {
+    const line = rawLine.trim();
+    if (line === "" || line.startsWith("%%")) continue;
+    if (!header && /^(flowchart|graph)\b/.test(line)) {
+      header = line;
+      continue;
+    }
+    let rest = line;
+    const firstM = rest.match(NODE_TOKEN);
+    if (!firstM || firstM[0] === "") continue;
+    let leftId = recordNode(nodes, firstM[0]);
+    rest = rest.slice(firstM[0].length).trimStart();
+    // Walk a (possibly chained) edge statement: A[..] --> B -->|x| C
+    while (rest.length > 0) {
+      const linkM = rest.match(LINK);
+      if (!linkM) break;
+      rest = rest.slice(linkM[0].length).trimStart();
+      const rhsM = rest.match(NODE_TOKEN);
+      if (!rhsM || rhsM[0] === "") break;
+      const rightId = recordNode(nodes, rhsM[0]);
+      edges.push({
+        from: leftId,
+        to: rightId,
+        label: linkM[3] || undefined,
+        index: edges.length,
+        raw: `${leftId} ${linkM[1]}${linkM[2] ?? ""} ${rightId}`,
+      });
+      rest = rest.slice(rhsM[0].length).trimStart();
+      leftId = rightId;
+    }
+  }
+  return { header: header || "flowchart TD", nodes, edges };
+}
+
+function nodeChanged(a: FlowNode, b: FlowNode): boolean {
+  return (a.label ?? "") !== (b.label ?? "") || (a.open ?? "") !== (b.open ?? "");
+}
+const edgeKey = (e: FlowEdge): string => `${e.from} ${e.to}`;
+
+export function diffFlowchart(beforeSrc: string, currentSrc: string): string {
+  const before = parseFlowchart(beforeSrc);
+  const current = parseFlowchart(currentSrc);
+
+  const addedNodes: string[] = [];
+  const changedNodes: string[] = [];
+  for (const [id, cur] of current.nodes) {
+    const prev = before.nodes.get(id);
+    if (!prev) addedNodes.push(id);
+    else if (nodeChanged(prev, cur)) changedNodes.push(id);
+  }
+  const removedNodes: FlowNode[] = [];
+  for (const [id, prev] of before.nodes) {
+    if (!current.nodes.has(id)) removedNodes.push(prev);
+  }
+
+  const beforeEdgeKeys = new Set(before.edges.map(edgeKey));
+  const currentEdgeKeys = new Set(current.edges.map(edgeKey));
+  const addedEdgeIdx = current.edges.filter((e) => !beforeEdgeKeys.has(edgeKey(e))).map((e) => e.index);
+  const removedEdges = before.edges.filter((e) => !currentEdgeKeys.has(edgeKey(e)));
+
+  // Ghost-edge indices follow the current edge count, in deterministic order.
+  let nextIdx = current.edges.length;
+  const ghostEdgeLines: string[] = [];
+  const ghostEdgeIdx: number[] = [];
+  for (const e of removedEdges) {
+    ghostEdgeLines.push(`  ${e.from} -.-> ${e.to}`);
+    ghostEdgeIdx.push(nextIdx++);
+  }
+
+  const out: string[] = [currentSrc.replace(/\s+$/, "")];
+
+  // Ghost removed nodes: re-inject their declaration (or bare id) so they appear.
+  for (const n of removedNodes) out.push(`  ${n.decl ?? n.id}`);
+  out.push(...ghostEdgeLines);
+
+  // classDefs (always emit the three; harmless if a class is unused).
+  out.push(
+    `  classDef cwAdded fill:${CW_COLORS.added}22,stroke:${CW_COLORS.added},stroke-width:2px;`,
+    `  classDef cwChanged fill:${CW_COLORS.changed}22,stroke:${CW_COLORS.changed},stroke-width:2px;`,
+    `  classDef cwRemoved fill:${CW_COLORS.removed}11,stroke:${CW_COLORS.removed},stroke-width:1px,stroke-dasharray:5 3,color:${CW_COLORS.removed};`,
+  );
+  if (addedNodes.length) out.push(`  class ${addedNodes.join(",")} cwAdded;`);
+  if (changedNodes.length) out.push(`  class ${changedNodes.join(",")} cwChanged;`);
+  if (removedNodes.length) out.push(`  class ${removedNodes.map((n) => n.id).join(",")} cwRemoved;`);
+
+  for (const i of addedEdgeIdx) out.push(`  linkStyle ${i} stroke:${CW_COLORS.added},stroke-width:2px;`);
+  for (const i of ghostEdgeIdx)
+    out.push(`  linkStyle ${i} stroke:${CW_COLORS.removed},stroke-width:1px,stroke-dasharray:5 3;`);
+
+  return out.join("\n");
+}
@@ -0,0 +1,103 @@
+/**
+ * mermaidSequenceDiff — F7.1 (#22). Pure sequence-diagram parser + diff/emit.
+ * Mermaid sequence diagrams have NO per-message color directive; the only
+ * per-message hook is the `rect rgb(...) … end` background block. So the emitter
+ * rebuilds the message stream (ghosted-removed messages re-inserted at their
+ * baseline position) wrapping each added/changed/removed message in a one-message
+ * tinted rect, and re-declares removed participants (INV-29/31). Deterministic;
+ * no vscode, no DOM.
+ */
+import { diffArrays } from "diff";
+import { CW_COLORS } from "./mermaidDiff";
+
+export interface SeqDiagram {
+  header: string;
+  participants: string[];
+  /** Message/other statement lines, verbatim & trimmed, in order. */
+  statements: string[];
+}
+
+// `A->>B: text`, plus ->, -->, -->>, -x, --x, -), --) variants. Captures from/to.
+const MSG = /^([A-Za-z0-9_]+)\s*(-{1,2}(?:>>?|x|\)))\s*([A-Za-z0-9_]+)\s*:/;
+const PARTICIPANT = /^(?:participant|actor)\s+([A-Za-z0-9_]+)/;
+
+export function parseSequence(source: string): SeqDiagram {
+  const lines = source.split(/\r?\n/);
+  let header = "";
+  const declared: string[] = [];
+  const seen = new Set<string>();
+  const statements: string[] = [];
+  const addP = (p: string) => {
+    if (!seen.has(p)) {
+      seen.add(p);
+      declared.push(p);
+    }
+  };
+  for (const rawLine of lines) {
+    const line = rawLine.trim();
+    if (line === "" || line.startsWith("%%")) continue;
+    if (!header && /^sequenceDiagram\b/.test(line)) {
+      header = line;
+      continue;
+    }
+    const pm = line.match(PARTICIPANT);
+    if (pm) {
+      addP(pm[1]);
+      continue;
+    }
+    const mm = line.match(MSG);
+    if (mm) {
+      addP(mm[1]);
+      addP(mm[3]);
+    }
+    statements.push(line);
+  }
+  return { header: header || "sequenceDiagram", participants: declared, statements };
+}
+
+function hexToRgb(hex: string): string {
+  const h = hex.replace("#", "");
+  const n = parseInt(h, 16);
+  return `${(n >> 16) & 255}, ${(n >> 8) & 255}, ${n & 255}`;
+}
+
+function rectWrap(stmt: string, hex: string): string[] {
+  return [`  rect rgb(${hexToRgb(hex)})`, `    ${stmt}`, `  end`];
+}
+
+export function diffSequence(beforeSrc: string, currentSrc: string): string {
+  const before = parseSequence(beforeSrc);
+  const current = parseSequence(currentSrc);
+
+  const out: string[] = [current.header];
+
+  // Participants: keep current declarations, then re-declare removed ones (ghosts).
+  for (const p of current.participants) out.push(`  participant ${p}`);
+  const curSet = new Set(current.participants);
+  for (const p of before.participants) {
+    if (!curSet.has(p)) out.push(`  participant ${p}`);
+  }
+
+  // Message stream diff via LCS over statements; pair adjacent removed+added as changed.
+  const parts = diffArrays(before.statements, current.statements);
+  for (let n = 0; n < parts.length; n++) {
+    const ch = parts[n];
+    if (!ch.added && !ch.removed) {
+      for (const s of ch.value) out.push(`  ${s}`);
+      continue;
+    }
+    if (ch.removed) {
+      const next = parts[n + 1];
+      const addVals = next?.added ? next.value : [];
+      const paired = Math.min(ch.value.length, addVals.length);
+      for (let k = 0; k < paired; k++) out.push(...rectWrap(addVals[k], CW_COLORS.changed));
+      for (let k = paired; k < ch.value.length; k++) out.push(...rectWrap(ch.value[k], CW_COLORS.removed));
+      for (let k = paired; k < addVals.length; k++) out.push(...rectWrap(addVals[k], CW_COLORS.added));
+      if (next?.added) n++;
+      continue;
+    }
+    // lone added run
+    for (const s of ch.value) out.push(...rectWrap(s, CW_COLORS.added));
+  }
+  return out.join("\n");
+}
@@ -90,6 +90,21 @@ export interface Proposal {
  turnId?: string;
  /** what the human asked for (review context). */
  instruction?: string;
+  /**
+   * F12/#47 (INV-39/40): the review-decision unit this proposal represents.
+   * `"block"` ⇒ the anchor spans a whole document block and accept reconciles
+   * attribution per WORD inside it (INV-40); `"single"` (or absent, for
+   * back-compat with older sidecars) ⇒ a single-range proposal accepted whole.
+   */
+  granularity?: "block" | "single";
+  /**
+   * F12/#64 (INV-48): the pre-apply text, captured when a proposal is
+   * optimistically applied to the buffer. `replacement` is now in the buffer and
+   * `fp.text` re-anchors to it, so `original` is the only record of what to revert
+   * to (revert-in-place) and what to show struck in the `<del>` half. Absent on a
+   * proposal created but not yet optimistically applied (or older sidecars).
+   */
+  original?: string;
 }

 export interface Artifact {
@@ -245,6 +260,8 @@ export function serializeArtifact(a: Artifact): string {
            createdAt: p.createdAt,
            ...(p.turnId !== undefined ? { turnId: p.turnId } : {}),
            ...(p.instruction !== undefined ? { instruction: p.instruction } : {}),
+            ...(p.original !== undefined ? { original: p.original } : {}),
+            ...(p.granularity !== undefined ? { granularity: p.granularity } : {}),
          },
          p,
        ),
@@ -0,0 +1,300 @@
+/**
+ * previewAnnotations — Task 7 (D3/D21, PUC-3): a pure, vscode-free markdown-it
+ * plugin that annotates the BUILT-IN VS Code Markdown preview with the F10
+ * authorship/change-mark vocabulary (`.cw-ins-claude` / `.cw-ins-human` /
+ * `.cw-del`) — the native-surfaces migration's replacement for the sunset
+ * webview's review body (spec §6.4/§6.10). Reuses the shipped sentinel
+ * discipline (`injectSentinels`/`sentinelsToSpans`, #33/#47-hardened) instead
+ * of inventing a new one: `annotateSource` computes PUA-sentinel-wrapped
+ * markdown SOURCE (never HTML) against the F6/F7 baseline + F3 spans + F4
+ * pending proposals; `cowritingMarkdownItPlugin` wires that transform into the
+ * shared `markdown-it` instance (the SAME instance VS Code's built-in preview
+ * uses via `markdown.markdownItPlugins`) as (a) a core rule — swap `state.src`
+ * before `normalize` — and (b) a full-render wrapper that turns surviving
+ * sentinel pairs into `<span class="cw-…">`.
+ *
+ * Implementation note (resolved drift from the migration plan's Task 7 sketch,
+ * which described (b) as "a text renderer rule"): `sentinelsToSpans` is a
+ * stateful, single-pass walk over one HTML string — installing it as a
+ * PER-TOKEN `renderer.rules.text` hook would reset that state at every text
+ * token boundary and silently break the "survives emphasis/tag boundaries"
+ * invariant (#33 CASE3) for any span whose sentinel-open and sentinel-close
+ * land in DIFFERENT text tokens (e.g. either side of a `**bold**` run — the
+ * exact scenario CASE3 exists to cover). Wrapping `md.renderer.render` instead
+ * runs the walk ONCE over the fully-assembled HTML, exactly like the F9/F10
+ * `colorByAuthor` precedent (`render(injected)` — a whole-block HTML string,
+ * not a token fragment) — this is what actually delivers the stated invariant,
+ * robust to callers that invoke `md.render()` directly or `md.parse()` +
+ * `md.renderer.render()` separately (VS Code's own scroll-sync `data-line`
+ * injection uses the latter).
+ *
+ * `host.inputsFor(env)` is the ONE thin, vscode-touching seam (owned by
+ * `extension.ts`): `env.currentDocument` is an OBSERVED VS Code preview-engine
+ * behavior, not a documented contract, so the host falls back to the last
+ * actively-coedited document when it is absent (see extension.ts's
+ * `lastCoeditedUri`).
+ *
+ * Deliberate choice: the mermaid-fence rendering (the `options.highlight`
+ * override installed below) applies to EVERY markdown preview, coedited or
+ * not — it is not gated on coediting state. Gating it would make a diagram's
+ * rendering flip (mermaid <-> plain fence) the instant a doc enters or exits
+ * coediting, which is a worse surprise than always rendering mermaid diagrams
+ * as diagrams; only the change-tracking OVERLAY on top (`buildMermaidQueue`'s
+ * diff augmentation) is gated on having a baseline to diff against.
+ */
+import type MarkdownIt from "markdown-it";
+import { diffMermaid } from "./mermaidDiff";
+import {
+  diffBlocks,
+  injectSentinels,
+  landedTextOf,
+  MERMAID_LEGEND,
+  mermaidFenceBody,
+  sentinelsToSpans,
+  splitBlocksWithRanges,
+  wordEditHunks,
+  type AuthorSpan,
+  type ProposalView,
+  type TaggedSpan,
+} from "./trackChangesModel";
+
+export interface AnnotationInputs {
+  baselineText: string | undefined;
+  baselineReason: "entered" | "pinned" | "head" | undefined;
+  spans: AuthorSpan[];
+  proposals: ProposalView[];
+  enabled: boolean;
+}
+
+/**
+ * Inject the sentinel vocabulary into `text` block-by-block (INV-23: code/
+ * mermaid fences stay atomic — never sentinel-injected, never word-refined) so
+ * the delimiter-run + block-marker safety `injectSentinels` carries applies
+ * PER BLOCK, not just at the very top of the whole document (its clamps are
+ * anchored to a block's own start, matching every other caller — `colorByAuthor`
+ * is always invoked per block too).
+ */
+function injectAllSentinels(text: string, marks: TaggedSpan[]): string {
+  if (marks.length === 0) return text;
+  const blocks = splitBlocksWithRanges(text);
+  if (blocks.length === 0) return text;
+  let out = "";
+  let cursor = 0;
+  for (const b of blocks) {
+    out += text.slice(cursor, b.start); // preserve inter-block gaps (blank lines) verbatim
+    if (b.type === "prose") {
+      const blockMarks = marks.filter((m) => m.end > b.start && m.start < b.end);
+      out += blockMarks.length ? injectSentinels(b.raw, b.start, blockMarks) : b.raw;
+    } else {
+      out += b.raw; // INV-23: fences are never annotated
+    }
+    cursor = b.end;
+  }
+  out += text.slice(cursor); // trailing content after the last block
+  return out;
+}
+
+/**
+ * Pure: markdown source -> sentinel-annotated source (authorship + change +
+ * proposal marks vs the F6/F7 baseline). Skips everything when annotations are
+ * disabled, or (the #48 rule) once the baseline was just PINNED with zero
+ * LANDED diff — pending F4 proposals are unaffected by that gate (matching the
+ * shipped preview's "pin→clean, proposals still show", #48/INV-33).
+ */
+export function annotateSource(src: string, inputs: AnnotationInputs): string {
+  if (!inputs.enabled) return src;
+  const proposals = inputs.proposals ?? [];
+  const baselineText = inputs.baselineText ?? src; // no baseline → no change-marks
+  const landedText = landedTextOf(src, proposals);
+  const pinnedZeroDiff = inputs.baselineReason === "pinned" && landedText === baselineText;
+
+  const insSpans: TaggedSpan[] = [];
+  const deletions: { at: number; text: string }[] = [];
+  const appliedRanges: { start: number; end: number }[] = [];
+
+  // (c) pending F4 proposals already sit applied in `src` (F12 optimistic
+  // apply, INV-48) — mark the applied range ins-<author> (`p.author` — a
+  // proposal defaults to "claude" per `ProposalView.author`, but a human can
+  // also be the proposer, e.g. a human-authored suggestion routed through the
+  // same F4 pending-proposal path) and resurface the pre-apply original as a
+  // deletion at the same point.
+  for (const p of proposals) {
+    if (p.anchorStart === null || p.anchorEnd === null) continue;
+    appliedRanges.push({ start: p.anchorStart, end: p.anchorEnd });
+    if (p.anchorEnd > p.anchorStart) insSpans.push({ start: p.anchorStart, end: p.anchorEnd, tag: p.author ?? "claude" });
+    const original = p.original ?? p.replaced;
+    if (original) deletions.push({ at: p.anchorStart, text: original });
+  }
+
+  // (a)/(b) committed changes-since-baseline. Each hunk (wordEditHunks(src,
+  // baseline) — start/end in SRC coords, `replacement` = the baseline-side text
+  // for that span) yields (a) insertion marks from the F3 `spans` CLIPPED to the
+  // hunk's changed range — a hunk may straddle more than one author (e.g. two
+  // adjacent words typed by different authors with nothing committed between
+  // them), so spans are clipped individually, never the whole hunk tagged with
+  // one author — and (b) the baseline's dropped text reinserted at the hunk
+  // start. A hunk owned by an applied pending proposal is already marked above.
+  if (!pinnedZeroDiff && landedText !== baselineText) {
+    for (const h of wordEditHunks(src, baselineText)) {
+      if (appliedRanges.some((r) => h.start < r.end && h.end > r.start)) continue;
+      for (const s of inputs.spans) {
+        const start = Math.max(s.start, h.start);
+        const end = Math.min(s.end, h.end);
+        if (end > start) insSpans.push({ start, end, tag: s.author });
+      }
+      if (h.replacement) deletions.push({ at: h.start, text: h.replacement });
+    }
+  }
+
+  if (insSpans.length === 0 && deletions.length === 0) return src;
+
+  // Splice deletion reinsertions into the source high→low (so earlier offsets
+  // stay valid), shifting every mark recorded so far and appending a "del" mark
+  // over the freshly-inserted text.
+  const marks: TaggedSpan[] = insSpans.map((s) => ({ ...s }));
+  let text = src;
+  for (const d of [...deletions].sort((a, b) => b.at - a.at)) {
+    text = text.slice(0, d.at) + d.text + text.slice(d.at);
+    for (const m of marks) {
+      if (m.start >= d.at) m.start += d.text.length;
+      if (m.end >= d.at) m.end += d.text.length;
+    }
+    marks.push({ start: d.at, end: d.at + d.text.length, tag: "del" });
+  }
+
+  return injectAllSentinels(text, marks);
+}
+
+/** One `buildMermaidQueue` entry: a changed mermaid fence's own body (for
+ * matching, `mermaidFenceBody` normalization — no trailing newline) paired
+ * with its `diffMermaid`-augmented replacement source. `consumed` is reset by
+ * the `render` wrapper (see `cowritingMarkdownItPlugin`) so the SAME entries
+ * can back a re-render without a fresh parse. */
+interface MermaidQueueEntry {
+  readonly curBody: string;
+  readonly source: string;
+  consumed: boolean;
+}
+
+/**
+ * Per-render queue of mermaid-fence overrides: one entry per `changed` atomic
+ * mermaid op (Task 7 §2.6 parity: reuses `diffBlocks`' baseline/current block
+ * pairing, the SAME pairing `renderReview`/`renderOp` used in the sunset
+ * webview, so a changed mermaid fence is re-emitted through `diffMermaid`
+ * exactly as it was there — INV-23 fences stay atomic, INV-29..31 styling
+ * directives) that has a diffable before/current pair AND a successful
+ * `diffMermaid` augmentation. `[]` when there is no baseline to diff against,
+ * annotations are disabled, or the landed text equals the baseline (covers
+ * the #48 pinned-zero-diff case too, since `diffBlocks` on identical text
+ * yields no `changed` ops).
+ *
+ * Diffed against `landedText` (proposals reverted), matching the "committed
+ * change since baseline" boundary `annotateSource`'s own `wordEditHunks(src,
+ * baselineText)` gate already draws (a pending, not-yet-accepted proposal
+ * isn't a landed change) — a fence a pending proposal's anchor overlaps is
+ * left unaugmented, same as the rest of that gate.
+ *
+ * NOT positional: `src` (what markdown-it actually parses fences from) and
+ * `landedText` (what this function diffs) can disagree on mermaid-fence COUNT
+ * or ORDER — e.g. a pending proposal optimistically applied in `src` (F12,
+ * INV-48) inserts a whole new mermaid fence that `landedText` (proposals
+ * reverted) doesn't have, or `annotateSource`'s committed-deletion reinsert
+ * splices an extra fence into the rendered source. `options.highlight`
+ * (`cowritingMarkdownItPlugin`) matches each queue entry to a fence by BODY,
+ * not by position — a fence with no matching entry renders verbatim (the
+ * queue is a bag of `changed` fences to find, not a strict per-fence
+ * override list), so misalignment degrades to verbatim exactly as intended,
+ * never substitutes the wrong diagram's augmentation into the wrong fence.
+ *
+ * Pure, vscode-free (INV-6).
+ */
+function buildMermaidQueue(src: string, inputs: AnnotationInputs): MermaidQueueEntry[] {
+  if (!inputs.enabled || inputs.baselineText === undefined) return [];
+  const baselineText = inputs.baselineText;
+  const landedText = landedTextOf(src, inputs.proposals ?? []);
+  if (landedText === baselineText) return [];
+  const entries: MermaidQueueEntry[] = [];
+  for (const op of diffBlocks(baselineText, landedText)) {
+    if (op.kind !== "changed" || op.block.type !== "mermaid") continue; // unchanged/added/removed: no override needed
+    const curBody = mermaidFenceBody(op.block.raw);
+    const beforeBody = op.before.type === "mermaid" ? mermaidFenceBody(op.before.raw) : null;
+    if (curBody === null || beforeBody === null) continue;
+    const result = diffMermaid(beforeBody, curBody);
+    if (result.kind === "augmented") entries.push({ curBody, source: result.source, consumed: false }); // INV-30 fallback: verbatim
+  }
+  return entries;
+}
+
+/**
+ * markdown-it plugin factory: the host supplies `AnnotationInputs` per render
+ * (`host.inputsFor(env)`; `undefined` = no coediting context for this render —
+ * source passed through untouched). Installs (a) a core rule BEFORE `normalize`
+ * that swaps `state.src` for `annotateSource(...)`, and (b) a full-render
+ * wrapper (see module docstring) that turns surviving sentinel pairs into
+ * `cw-…` spans. Also teaches ```mermaid fences to render as `<pre class="mermaid">`
+ * (Q4/Task 7 Step 2.6) via `options.highlight` — the same extension point the
+ * built-in preview's own fence rule already calls for syntax highlighting — so
+ * the bundled `media/preview-mermaid.js` previewScript has something to run
+ * mermaid over (the built-in preview does not render mermaid on its own). A
+ * CHANGED mermaid fence's source is re-emitted through `diffMermaid` first
+ * (`buildMermaidQueue`, Task 7 §2.6 parity) so mermaid renders the intra-
+ * diagram diff (INV-29..31), not just the current diagram. `options.highlight`
+ * has no `env`/`state` parameter of its own, so the per-render queue is
+ * threaded through a closure variable set by the core rule (which DOES see
+ * `state.env`) and consumed BY BODY MATCH, not fence position (see
+ * `buildMermaidQueue`), by `highlight`: each fence's `code` is matched against
+ * unconsumed queue entries in order, the first match is consumed (so a later
+ * fence with the same body still finds its own entry), and a fence with no
+ * match renders verbatim WITHOUT consuming anything (so a later fence can
+ * still match). markdown-it's `parse` (core rules) always completes before
+ * `renderer.render` invokes `highlight`, and a single md instance never
+ * renders concurrently, so the closure handoff itself is safe — but VS Code
+ * may call `md.parse()` and `md.renderer.render()` separately and re-render
+ * cached tokens without a fresh parse (the same caller pattern the module
+ * docstring notes for scroll-sync), which would otherwise replay a queue whose
+ * entries were already marked consumed by the prior render; the `render`
+ * wrapper resets every entry's `consumed` flag so each render pass starts with
+ * the full queue available again.
+ */
+export function cowritingMarkdownItPlugin(
+  md: MarkdownIt,
+  host: { inputsFor(env: unknown): AnnotationInputs | undefined },
+): MarkdownIt {
+  let mermaidQueue: MermaidQueueEntry[] = [];
+
+  md.core.ruler.before("normalize", "cowriting-annotate", (state) => {
+    const inputs = host.inputsFor(state.env);
+    mermaidQueue = inputs ? buildMermaidQueue(state.src, inputs) : [];
+    if (!inputs) return;
+    state.src = annotateSource(state.src, inputs);
+  });
+
+  const baseRender = md.renderer.render.bind(md.renderer);
+  md.renderer.render = (tokens, options, env) => {
+    for (const entry of mermaidQueue) entry.consumed = false; // re-render-without-reparse safety net
+    return sentinelsToSpans(baseRender(tokens, options, env), "ins");
+  };
+
+  const baseHighlight = md.options.highlight;
+  md.options.highlight = (code, lang, attrs) => {
+    if (lang?.trim().toLowerCase() === "mermaid") {
+      const normalized = code.replace(/\n+$/, ""); // token.content keeps a trailing LF; mermaidFenceBody doesn't
+      const entry = mermaidQueue.find((e) => !e.consumed && e.curBody === normalized);
+      if (entry) entry.consumed = true;
+      const legend = entry ? MERMAID_LEGEND : "";
+      return `<pre class="mermaid" style="all: unset;">${escapeMermaidSource(entry?.source ?? code)}</pre>${legend}`;
+    }
+    return baseHighlight?.(code, lang, attrs) ?? md.utils.escapeHtml(code);
+  };
+
+  return md;
+}
+
+/** Minimal HTML-escape for a mermaid fence body (no markdown parsing — verbatim diagram source). */
+function escapeMermaidSource(source: string): string {
+  return source
+    .replace(/&/g, "&amp;")
+    .replace(/</g, "&lt;")
+    .replace(/>/g, "&gt;")
+    .replace(/\n+$/, "");
+}
@@ -4,18 +4,22 @@
 * ingress (INV-10: NEVER mutates the document), persistence at propose time,
 * resolve-or-flag on load/external change (INV-11 — a proposal's anchor is
 * immutable for its life: no save-time re-fingerprint, unlike threads),
- * rendering (second Comments controller + amber pending-range decoration),
+ * anchor bookkeeping into state.live/state.unresolved (no in-editor UI —
+ * F10/INV-32 makes the rendered preview the single review surface),
 * and the human-only accept/reject gestures (INV-12). Accept drives the seam
 * (AttributionController.applyAgentEdit, INV-9) so accepted text lands
 * Claude-attributed with zero new attribution code.
 */
 import * as vscode from "vscode";
-import { CoauthorStore } from "./store";
+import { SidecarRouter, docIdentity } from "./sidecarRouter";
 import { emptyArtifact, type Artifact, type Fingerprint, type Proposal, type Provenance } from "./model";
-import { resolve, shift, type OffsetRange } from "./anchorer";
-import { addProposal, proposalBody, removeProposal } from "./proposalModel";
+import { resolve, shift, shiftTracked, buildFingerprint, type OffsetRange } from "./anchorer";
+import { addProposal, removeProposal, setProposalApplied } from "./proposalModel";
 import type { AttributionController } from "./attributionController";
 import type { VersionGuard } from "./versionGuard";
+import { isAuthorable } from "./workspacePath";
+import { wordEditHunks, type ProposalView } from "./trackChangesModel";
+import type { CoeditingRegistry } from "./coeditingRegistry";

 /** Test-facing snapshot of what is currently rendered for a document. */
 export interface RenderedProposal {
@@ -32,60 +36,123 @@ interface DocState {
  docPath: string;
  uri: vscode.Uri;
  artifact: Artifact;
-  vsThreads: Map<string, vscode.CommentThread>;
  /** proposal id -> live offset range (within-session optimization, INV-3). */
  live: Map<string, OffsetRange>;
  /** proposal ids whose anchor did not resolve at last render (stale/orphaned). */
  unresolved: Set<string>;
+  /** ids already optimistically applied to the buffer (so the trigger doesn't re-apply). */
+  applied: Set<string>;
+  /**
+   * #70 (INV-5): the F12 applied-range bookkeeping — proposal id -> the applied
+   * span's CURRENT buffer range. Written at optimistic-apply, re-synced whenever
+   * the exact anchor resolves (renderAll), shifted through interior-safe edits,
+   * and DELETED (distrusted) when an edit straddles a span boundary — a clamped
+   * range is a guess, and stale text is never applied by guess (INV-11). Lets
+   * Reject restore the original after the human types INSIDE the pending range
+   * (which orphans the exact-substring anchor).
+   */
+  appliedSpans: Map<string, OffsetRange>;
 }

-const PENDING_DECO: vscode.DecorationRenderOptions = {
-  backgroundColor: "rgba(245, 158, 11, 0.18)",
-  overviewRulerColor: "rgba(245, 158, 11, 0.8)",
-  overviewRulerLane: vscode.OverviewRulerLane.Right,
-};
-
 export class ProposalController implements vscode.Disposable {
-  private readonly controller: vscode.CommentController;
  private readonly disposables: vscode.Disposable[] = [];
  private readonly docs = new Map<string, DocState>(); // keyed by docPath
-  private readonly pendingType = vscode.window.createTextEditorDecorationType(PENDING_DECO);
  private readonly statusItem = vscode.window.createStatusBarItem(vscode.StatusBarAlignment.Right, 89);
+  private readonly onDidChangeProposalsEmitter = new vscode.EventEmitter<{ uri: string }>();
+  /** Fires on propose / accept / reject / external sidecar change (F10). */
+  readonly onDidChangeProposals = this.onDidChangeProposalsEmitter.event;

  constructor(
-    private readonly store: CoauthorStore,
+    private readonly store: SidecarRouter,
    private readonly attribution: AttributionController,
-    private readonly rootDir: string,
+    private readonly rootDir: string | undefined,
    private readonly guard: VersionGuard,
+    private readonly registry: CoeditingRegistry,
  ) {
-    // No commentingRangeProvider: humans never open proposal threads by hand —
-    // proposals are born of machine turns only (INV-12 keeps decisions human).
-    this.controller = vscode.comments.createCommentController("cowriting.proposals", "Claude Proposals");
-    this.disposables.push(this.controller, this.pendingType, this.statusItem);
+    this.disposables.push(this.statusItem);
+    this.disposables.push(this.onDidChangeProposalsEmitter);
    this.disposables.push(
-      vscode.commands.registerCommand("cowriting.acceptProposal", (t: vscode.CommentThread) => this.acceptThread(t)),
-      vscode.commands.registerCommand("cowriting.rejectProposal", (t: vscode.CommentThread) => this.rejectThread(t)),
      vscode.workspace.onDidChangeTextDocument((e) => this.onDidChange(e)),
+      // #70: a tracked applied span is only meaningful while its buffer is live —
+      // a closed doc can change on disk with no change events, so a span kept
+      // across close/reopen could revert over unrelated text (INV-11: never by
+      // guess). Cleared here; a resolving anchor rebuilds it at the next renderAll.
+      vscode.workspace.onDidCloseTextDocument((doc) => {
+        this.docs.get(this.keyOf(doc))?.appliedSpans.clear();
+      }),
+      // PUC-7 restore-on-enter (mirrors ThreadController/AttributionController):
+      // `renderAll` is the only path that recomputes state.live/unresolved AND
+      // fires onDidChangeProposals — the event EditorProposalController listens
+      // to for optimistic-apply (decorations/CodeLens). Without this, a FIRST
+      // entry into coediting (registry.enter()) never fires it, so pre-existing
+      // pending proposals stay un-applied/undecorated until the next edit.
+      // `renderAll` already gates on isCoediting; exit intentionally does
+      // nothing here (it clears only the live/unresolved recompute, not the
+      // persisted artifact — EditorProposalController's own gate hides the
+      // decorations). Registered after AttributionController's own subscription
+      // (construction order in extension.ts) so decorateCommitted's spansFor
+      // read sees fresh attribution on the same enter transition.
+      this.registry.onDidChange(({ uri, coediting }) => {
+        if (!coediting) {
+          // Finding 5 (final whole-branch review): exit hides the status-bar
+          // item too — otherwise a stale "N stale proposals" warning for a
+          // doc that's no longer even coediting stays pinned in the bar until
+          // some unrelated render happens to fire for it.
+          if (vscode.window.activeTextEditor?.document.uri.toString() === uri) this.statusItem.hide();
+          return;
+        }
+        const doc = vscode.workspace.textDocuments.find((d) => d.uri.toString() === uri);
+        if (doc) this.renderAll(doc);
+      }),
    );
  }

-  private isTracked(document: vscode.TextDocument): boolean {
-    return document.uri.scheme === "file" && document.uri.fsPath.startsWith(this.rootDir);
+  private fireChanged(document: vscode.TextDocument): void {
+    this.onDidChangeProposalsEmitter.fire({ uri: document.uri.toString() });
  }
-  private docPathOf(uri: vscode.Uri): string {
-    return vscode.workspace.asRelativePath(uri, false);
+
+  private isTracked(document: vscode.TextDocument): boolean {
+    return isAuthorable(document.uri.scheme);
+  }
+  /** The single document key (F8): repo-relative path in-workspace, URI string otherwise. */
+  private keyOf(document: vscode.TextDocument): string {
+    return this.store.keyOf(docIdentity(document));
+  }
+  /** The doc key F4 uses (F8 routing) — exposed for F10's preview. */
+  keyFor(document: vscode.TextDocument): string {
+    return this.keyOf(document);
+  }
+  /** Resolved proposal views for the F10 preview (anchorStart=null when unresolved). */
+  listProposals(document: vscode.TextDocument): ProposalView[] {
+    const docPath = this.keyOf(document);
+    const artifact = this.store.load(docPath) ?? emptyArtifact(docPath);
+    const text = document.getText();
+    return artifact.proposals.map((p) => {
+      const fp = artifact.anchors[p.anchorId]?.fingerprint;
+      const resolved = fp ? resolve(text, fp) : "orphaned";
+      return {
+        id: p.id,
+        anchorStart: resolved === "orphaned" ? null : resolved.start,
+        anchorEnd: resolved === "orphaned" ? null : resolved.end,
+        replaced: p.original ?? fp?.text ?? "",
+        replacement: p.replacement,
+        original: p.original,
+        author: p.author?.kind === "agent" ? "claude" : "human",
+      };
+    });
  }
  private ensureState(document: vscode.TextDocument): DocState {
-    const docPath = this.docPathOf(document.uri);
+    const docPath = this.keyOf(document);
    let state = this.docs.get(docPath);
    if (!state) {
      state = {
        docPath,
        uri: document.uri,
        artifact: this.store.load(docPath) ?? emptyArtifact(docPath),
-        vsThreads: new Map(),
        live: new Map(),
        unresolved: new Set(),
+        applied: new Set(),
+        appliedSpans: new Map(),
      };
      this.docs.set(docPath, state);
    }
@@ -104,11 +171,11 @@ export class ProposalController implements vscode.Disposable {
    fp: Fingerprint,
    replacement: string,
    author: Provenance,
-    opts?: { turnId?: string; instruction?: string },
+    opts?: { turnId?: string; instruction?: string; granularity?: "block" | "single" },
  ): Promise<string | undefined> {
    if (!this.isTracked(document)) return undefined;
-    if (this.guard.isReadOnly(this.docPathOf(document.uri))) return undefined;
-    const docPath = this.docPathOf(document.uri);
+    if (this.guard.isReadOnly(this.keyOf(document))) return undefined;
+    const docPath = this.keyOf(document);
    let proposalId: string | undefined;
    this.store.update(docPath, (a) => {
      proposalId = addProposal(a, fp, replacement, author, opts).proposalId;
@@ -119,10 +186,66 @@ export class ProposalController implements vscode.Disposable {

  // ---- PUC-2/PUC-3: accept / reject (INV-11/INV-12) ----------------------------------

-  /** Accept by proposal id (test-facing twin of the thread-menu gesture). */
-  async acceptById(docPath: string, proposalId: string): Promise<boolean> {
+  /** Accept by id — F12: finalize the already-applied text in place (INV-51). */
+  async acceptById(docPath: string, proposalId: string, opts?: { silent?: boolean }): Promise<boolean> {
+    if (this.isApplied(docPath, proposalId)) {
+      // INV-11 (applied path): if an external write mangled the optimistically-applied
+      // text so the fingerprint no longer resolves, refuse finalize — same guard as the
+      // legacy accept path.  Direct finalizeInPlace calls (CodeLens Accept gesture where
+      // the user may have edited inside the applied span) bypass this check intentionally.
+      const hit = this.byId(docPath, proposalId);
+      if (hit) {
+        const document = this.openDoc(hit.state);
+        if (document) {
+          const fp = hit.state.artifact.anchors[hit.proposal.anchorId]?.fingerprint;
+          const resolved = fp ? resolve(document.getText(), fp) : "orphaned";
+          if (resolved === "orphaned") {
+            if (!opts?.silent) {
+              void vscode.window.showWarningMessage(
+                "Cowriting: this proposal's target text changed or is missing — undo to restore it, or reject to discard (it is never applied by guess).",
+              );
+            }
+            return false;
+          }
+        }
+      }
+      return this.finalizeInPlace(docPath, proposalId);
+    }
+    // Fallback: a proposal that was never optimistically applied (e.g. orphaned at
+    // apply time) keeps the legacy seam-apply accept.
    const hit = this.byId(docPath, proposalId);
-    return hit ? this.accept(hit.state, hit.proposal) : false;
+    return hit ? this.accept(hit.state, hit.proposal, opts) : false;
+  }
+
+  /**
+   * #46 (INV-42): accept EVERY pending proposal on a document in one gesture — a
+   * batched application of the existing `acceptById` seam, not a new mechanism.
+   * Block proposals take the INV-40 word-precise path automatically. Applied in
+   * DESCENDING anchor order so an earlier accept never invalidates a later one's
+   * offsets; proposals whose anchor can't resolve are SKIPPED (never force-applied)
+   * and counted. Returns the applied-vs-skipped tally for the caller to report.
+   */
+  async acceptAllProposals(document: vscode.TextDocument): Promise<{ applied: number; skipped: number }> {
+    if (!this.isTracked(document)) return { applied: 0, skipped: 0 };
+    const state = this.ensureState(document);
+    state.artifact = this.store.load(state.docPath) ?? emptyArtifact(state.docPath);
+    const text = document.getText();
+    const items = state.artifact.proposals.map((p) => {
+      const fp = state.artifact.anchors[p.anchorId]?.fingerprint;
+      const resolved = fp ? resolve(text, fp) : "orphaned";
+      return { id: p.id, start: resolved === "orphaned" ? null : resolved.start };
+    });
+    const resolvable = items
+      .filter((i): i is { id: string; start: number } => i.start !== null)
+      .sort((a, b) => b.start - a.start);
+    let applied = 0;
+    let skipped = items.length - resolvable.length; // orphans, skipped up front
+    for (const it of resolvable) {
+      // silent: one batch report stands in for N per-proposal warnings.
+      if (await this.acceptById(state.docPath, it.id, { silent: true })) applied++;
+      else skipped++;
+    }
+    return { applied, skipped };
  }
  /** Reject by proposal id (test-facing twin of the thread-menu gesture). */
  rejectById(docPath: string, proposalId: string): boolean {
@@ -132,16 +255,13 @@ export class ProposalController implements vscode.Disposable {
    return true;
  }

-  private async acceptThread(vsThread: vscode.CommentThread): Promise<void> {
-    const hit = this.byThread(vsThread);
-    if (hit) await this.accept(hit.state, hit.proposal);
-  }
-  private rejectThread(vsThread: vscode.CommentThread): void {
-    const hit = this.byThread(vsThread);
-    if (hit) this.reject(hit.state, hit.proposal);
+  /** Reject by id — F12: revert the applied text in place (INV-51). */
+  async rejectByIdInPlace(docPath: string, proposalId: string): Promise<boolean> {
+    if (this.isApplied(docPath, proposalId)) return this.revertInPlace(docPath, proposalId);
+    return this.rejectById(docPath, proposalId);
  }

-  private async accept(state: DocState, proposal: Proposal): Promise<boolean> {
+  private async accept(state: DocState, proposal: Proposal, opts?: { silent?: boolean }): Promise<boolean> {
    if (this.guard.isReadOnly(state.docPath)) return false;
    const document = this.openDoc(state);
    if (!document) return false;
@@ -149,22 +269,32 @@ export class ProposalController implements vscode.Disposable {
    const fp = state.artifact.anchors[proposal.anchorId]?.fingerprint;
    const resolved = fp ? resolve(document.getText(), fp) : "orphaned";
    if (resolved === "orphaned") {
-      void vscode.window.showWarningMessage(
-        "Cowriting: this proposal's target text changed or is missing — undo to restore it, or reject to discard (it is never applied by guess).",
-      );
+      // #46: accept-all suppresses per-proposal warnings (one batch report instead).
+      if (!opts?.silent)
+        void vscode.window.showWarningMessage(
+          "Cowriting: this proposal's target text changed or is missing — undo to restore it, or reject to discard (it is never applied by guess).",
+        );
      this.renderAll(document);
      return false;
    }
-    const range = new vscode.Range(document.positionAt(resolved.start), document.positionAt(resolved.end));
-    // No awaits between resolve and the seam call: document.version is current.
-    const ok = await this.attribution.applyAgentEdit(document, range, proposal.replacement, proposal.author, {
-      expectedVersion: document.version,
-      turnId: proposal.turnId,
-    });
+    // #47 (INV-40): a BLOCK proposal applies the whole block but attributes only
+    // the words Claude actually changed; a single proposal applies its whole range.
+    const ok =
+      proposal.granularity === "block"
+        ? await this.acceptBlock(document, resolved, proposal)
+        : await this.attribution.applyAgentEdit(
+            document,
+            new vscode.Range(document.positionAt(resolved.start), document.positionAt(resolved.end)),
+            proposal.replacement,
+            proposal.author,
+            // No awaits between resolve and the seam call: document.version is current.
+            { expectedVersion: document.version, turnId: proposal.turnId },
+          );
    if (!ok) {
-      void vscode.window.showWarningMessage(
-        "Cowriting: the editor rejected the accept — the proposal is still pending.",
-      );
+      if (!opts?.silent)
+        void vscode.window.showWarningMessage(
+          "Cowriting: the editor rejected the accept — the proposal is still pending.",
+        );
      return false;
    }
    this.store.update(state.docPath, (a) => removeProposal(a, proposal.id));
@@ -172,25 +302,270 @@ export class ProposalController implements vscode.Disposable {
    return true;
  }

+  /**
+   * #47 (INV-40): accept a BLOCK proposal — apply Claude's whole block, but
+   * attribute ONLY the runs Claude actually changed. The block is the DECISION
+   * unit; the word is the ATTRIBUTION unit. An intra-block word sub-diff
+   * (`wordEditHunks`, the raw engine INV-37 used, repurposed — un-anchored so the
+   * runs stay disjoint for a batch apply) yields the changed runs; each lands
+   * through the F4 seam (Claude-attributed), applied last-position-first so an
+   * earlier run's offsets stay valid under the later ones. Unchanged spans within
+   * the block are never touched, so their prior authorship stands. (Each run is
+   * one seam edit / one undo step — see the spec's deferred note on undo grouping.)
+   */
+  private async acceptBlock(
+    document: vscode.TextDocument,
+    resolved: OffsetRange,
+    proposal: Proposal,
+  ): Promise<boolean> {
+    const blockText = document.getText(
+      new vscode.Range(document.positionAt(resolved.start), document.positionAt(resolved.end)),
+    );
+    const subHunks = wordEditHunks(blockText, proposal.replacement);
+    if (subHunks.length === 0) return true; // block already equals the proposal — nothing to attribute
+    for (const h of [...subHunks].sort((a, b) => b.start - a.start)) {
+      const range = new vscode.Range(
+        document.positionAt(resolved.start + h.start),
+        document.positionAt(resolved.start + h.end),
+      );
+      const ok = await this.attribution.applyAgentEdit(document, range, h.replacement, proposal.author, {
+        expectedVersion: document.version,
+        turnId: proposal.turnId,
+      });
+      if (!ok) return false;
+    }
+    return true;
+  }
+
+  /** True once this proposal's text is in the buffer (optimistic apply ran). */
+  isApplied(docPath: string, proposalId: string): boolean {
+    return this.docs.get(docPath)?.applied.has(proposalId) ?? false;
+  }
+
+  /** #70: the shift-tracked applied span (the revert-fallback bookkeeping) —
+   *  exposed so the editor surface can keep the decision gestures reachable
+   *  for an applied proposal whose exact anchor an interior tweak orphaned. */
+  trackedSpan(docPath: string, proposalId: string): OffsetRange | undefined {
+    return this.docs.get(docPath)?.appliedSpans.get(proposalId);
+  }
+
+  /** The one proposal-clear sequence (record + in-memory bookkeeping + re-render)
+   *  shared by finalize/revert/reject so no path leaves a stale entry behind. */
+  private clearProposal(state: DocState, proposalId: string, document?: vscode.TextDocument): void {
+    this.store.update(state.docPath, (a) => removeProposal(a, proposalId));
+    state.applied.delete(proposalId);
+    state.appliedSpans.delete(proposalId);
+    if (document) this.renderAll(document);
+  }
+
+  /**
+   * F12/#64 (INV-48): optimistically apply a pending proposal INTO the buffer so
+   * the editor shows the would-be-accepted result (editable). Reuses the F4
+   * word-precise seam (block → per-word hunks, INV-40; single → whole range) but
+   * with `landBaseline:false` (the change stays pending). Then re-anchors the
+   * proposal to the applied text and stores the original (`setProposalApplied`), so
+   * `resolve()` finds it in the mutated buffer and revert/decorate key off it.
+   * Idempotent: a no-op if already applied.
+   */
+  async optimisticApply(document: vscode.TextDocument, proposalId: string): Promise<boolean> {
+    if (!this.isTracked(document) || this.guard.isReadOnly(this.keyOf(document))) return false;
+    const docPath = this.keyOf(document);
+    const state = this.ensureState(document);
+    if (state.applied.has(proposalId)) return true;
+    state.artifact = this.store.load(docPath) ?? emptyArtifact(docPath);
+    const proposal = state.artifact.proposals.find((p) => p.id === proposalId);
+    const fp = proposal ? state.artifact.anchors[proposal.anchorId]?.fingerprint : undefined;
+    if (!proposal || !fp) return false;
+    // Reload-safety (INV-51/54): a proposal that already carries `original` was
+    // optimistically applied in a PRIOR session — the buffer holds the applied text
+    // and `fp` points at it, but this (fresh) controller's in-memory `applied` set is
+    // empty. Re-applying would recapture `original` from the already-applied buffer
+    // (= the replacement) and CLOBBER the true revert target, breaking Reject. Mark it
+    // applied in memory and stop — `original` is captured exactly once, on first apply.
+    if (proposal.original !== undefined) {
+      state.applied.add(proposalId);
+      this.renderAll(document);
+      return true;
+    }
+    const resolved = resolve(document.getText(), fp);
+    if (resolved === "orphaned") return false;
+    const original = document.getText(
+      new vscode.Range(document.positionAt(resolved.start), document.positionAt(resolved.end)),
+    );
+    const ok =
+      proposal.granularity === "block"
+        ? await this.applyBlockOptimistic(document, resolved, proposal)
+        : await this.attribution.applyAgentEdit(
+            document,
+            new vscode.Range(document.positionAt(resolved.start), document.positionAt(resolved.end)),
+            proposal.replacement,
+            proposal.author,
+            { expectedVersion: document.version, turnId: proposal.turnId, landBaseline: false },
+          );
+    if (!ok) return false;
+    state.applied.add(proposalId);
+    // Re-anchor to the applied text now in the buffer (its start is unchanged; its
+    // end shifts by the net length delta of the replacement).
+    const appliedStart = resolved.start;
+    const appliedEnd = appliedStart + proposal.replacement.length;
+    state.appliedSpans.set(proposalId, { start: appliedStart, end: appliedEnd });
+    const appliedFp = buildFingerprint(document.getText(), { start: appliedStart, end: appliedEnd });
+    this.store.update(docPath, (a) => setProposalApplied(a, proposalId, appliedFp, original));
+    this.renderAll(document);
+    return true;
+  }
+
+  /** Block optimistic apply: the INV-40 per-word hunks, but landBaseline:false. */
+  private async applyBlockOptimistic(
+    document: vscode.TextDocument,
+    resolved: OffsetRange,
+    proposal: Proposal,
+  ): Promise<boolean> {
+    const blockText = document.getText(
+      new vscode.Range(document.positionAt(resolved.start), document.positionAt(resolved.end)),
+    );
+    const subHunks = wordEditHunks(blockText, proposal.replacement);
+    if (subHunks.length === 0) return true;
+    for (const h of [...subHunks].sort((a, b) => b.start - a.start)) {
+      const range = new vscode.Range(
+        document.positionAt(resolved.start + h.start),
+        document.positionAt(resolved.start + h.end),
+      );
+      const ok = await this.attribution.applyAgentEdit(document, range, h.replacement, proposal.author, {
+        expectedVersion: document.version, turnId: proposal.turnId, landBaseline: false,
+      });
+      if (!ok) return false;
+    }
+    return true;
+  }
+
+  /**
+   * F12/#64 (INV-51): ACCEPT an optimistically-applied proposal — the text is
+   * already in the buffer, so this only marks the attribution landing (via
+   * `attribution.signalLanded`) and clears the proposal. No re-application.
+   * Native-surfaces migration (INV-7/INV-18 retirement, spec §6.4): this no
+   * longer advances the baseline — a landed proposal stays a visible change-
+   * since-baseline until commit (head mode) or "Mark Changes as Reviewed"
+   * (snapshot mode); nothing here subscribes to `signalLanded` for baseline
+   * purposes anymore (the `onDidApplyAgentEdit → advance` wire was removed).
+   */
+  async finalizeInPlace(docPath: string, proposalId: string): Promise<boolean> {
+    if (this.guard.isReadOnly(docPath)) return false;
+    const hit = this.byId(docPath, proposalId);
+    if (!hit) return false;
+    const document = this.openDoc(hit.state);
+    if (!document) return false;
+    this.attribution.signalLanded(document);
+    this.clearProposal(hit.state, proposalId, document);
+    return true;
+  }
+
+  /**
+   * F12/#64 (INV-51): REJECT an optimistically-applied proposal — replace its live
+   * applied span with the stored `original`, then clear it. Reverts the whole block
+   * regardless of any in-place edits the human made to the inserted text: when an
+   * interior tweak has orphaned the exact anchor (INV-11), the revert falls back to
+   * the shift-tracked applied span (#70, INV-5). When neither locates the span
+   * (e.g. a boundary-straddling external rewrite distrusted it), the reject FAILS
+   * loudly — warning shown, proposal kept, `false` returned — instead of silently
+   * leaving the proposed text in the buffer. Undo (or ✓ Keep) remains the way out.
+   */
+  async revertInPlace(docPath: string, proposalId: string, opts?: { silent?: boolean }): Promise<boolean> {
+    if (this.guard.isReadOnly(docPath)) return false;
+    const hit = this.byId(docPath, proposalId);
+    if (!hit) return false;
+    const document = this.openDoc(hit.state);
+    if (!document) return false;
+    // Never optimistically applied → nothing of ours is in the buffer; clearing
+    // the record IS the reject (the legacy pending-only path).
+    if (hit.proposal.original === undefined) {
+      this.clearProposal(hit.state, proposalId, document);
+      return true;
+    }
+    const fp = hit.state.artifact.anchors[hit.proposal.anchorId]?.fingerprint;
+    const resolved = fp ? resolve(document.getText(), fp) : "orphaned";
+    const span = resolved !== "orphaned" ? resolved : hit.state.appliedSpans.get(proposalId);
+    if (!span) {
+      if (!opts?.silent) {
+        // The revert is refused, but the record must stay dismissable (a window
+        // reload empties both the tracked spans and the undo stack): offer a
+        // record-only discard that leaves the buffer exactly as it stands.
+        const DISCARD = "Discard proposal (leave text)";
+        void vscode.window
+          .showWarningMessage(
+            "Cowriting: this proposal's applied text can't be located (it changed or moved) — it is never reverted by guess. Undo your edits, Keep it, or discard the record as-is.",
+            DISCARD,
+          )
+          .then((choice) => {
+            if (choice === DISCARD) this.clearProposal(hit.state, proposalId, document);
+          });
+      }
+      return false;
+    }
+    const we = new vscode.WorkspaceEdit();
+    we.replace(
+      document.uri,
+      new vscode.Range(document.positionAt(span.start), document.positionAt(span.end)),
+      hit.proposal.original,
+    );
+    if (!(await vscode.workspace.applyEdit(we))) return false;
+    this.clearProposal(hit.state, proposalId, document);
+    return true;
+  }
+
+  /**
+   * F12/#64 (INV-53): reject EVERY pending proposal on a document — revert each in
+   * DESCENDING span order (so an earlier revert never shifts a later one's
+   * offsets), symmetric with #46's accept-all. #70: a tweaked proposal whose exact
+   * anchor is orphaned orders (and reverts) by its shift-tracked applied span;
+   * one with no locatable span is SKIPPED — counted, never guessed (INV-11) —
+   * and the per-proposal warning is suppressed in favor of the batch report.
+   */
+  async rejectAll(document: vscode.TextDocument): Promise<{ reverted: number; skipped: number }> {
+    if (!this.isTracked(document)) return { reverted: 0, skipped: 0 };
+    const docPath = this.keyOf(document);
+    const state = this.ensureState(document);
+    state.artifact = this.store.load(docPath) ?? emptyArtifact(docPath);
+    const text = document.getText();
+    const ordered = state.artifact.proposals
+      .map((p) => {
+        const fp = state.artifact.anchors[p.anchorId]?.fingerprint;
+        const r = fp ? resolve(text, fp) : "orphaned";
+        const start = r !== "orphaned" ? r.start : state.appliedSpans.get(p.id)?.start ?? -1;
+        return { id: p.id, start };
+      })
+      .sort((a, b) => b.start - a.start);
+    let reverted = 0;
+    let skipped = 0;
+    for (const it of ordered) {
+      if (await this.revertInPlace(docPath, it.id, { silent: true })) reverted++;
+      else skipped++;
+    }
+    return { reverted, skipped };
+  }
+
  private reject(state: DocState, proposal: Proposal): void {
    if (this.guard.isReadOnly(state.docPath)) return;
-    this.store.update(state.docPath, (a) => removeProposal(a, proposal.id));
-    const document = this.openDoc(state);
-    if (document) this.renderAll(document);
+    this.clearProposal(state, proposal.id, this.openDoc(state));
  }

  // ---- PUC-4: load / external change / resolve-or-flag --------------------------------

-  /** Load + (re)render every pending proposal at its resolved anchor (or flagged). */
+  /** Load + (re)render every pending proposal at its resolved anchor (or flagged).
+   *  INV-10: the RENDER portion (live/unresolved recompute, the status bar, and
+   *  the onDidChangeProposals fire that drives F12's optimistic-apply/CodeLens/
+   *  decorations) is gated on coediting — `propose()` itself stays ungated (it
+   *  always persists), and so does the state LOAD below: `acceptById`/`rejectById`
+   *  key off `this.docs` (populated here) regardless of coediting, since deciding
+   *  an already-persisted proposal is not a "rendering" surface. */
  renderAll(document: vscode.TextDocument): void {
    if (!this.isTracked(document)) return;
-    const docPath = this.docPathOf(document.uri);
+    const docPath = this.keyOf(document);
    const state = this.ensureState(document);
    state.artifact = this.store.load(docPath) ?? emptyArtifact(docPath);
-    for (const vsThread of state.vsThreads.values()) vsThread.dispose();
-    state.vsThreads.clear();
    state.live.clear();
    state.unresolved.clear();
+    if (!this.registry.isCoediting(document.uri)) return;
    const text = document.getText();
    for (const proposal of state.artifact.proposals) {
      const fp = state.artifact.anchors[proposal.anchorId]?.fingerprint;
@@ -198,90 +573,64 @@ export class ProposalController implements vscode.Disposable {
      if (resolved === "orphaned") {
        const line = fp ? Math.min(fp.lineHint, Math.max(0, document.lineCount - 1)) : 0;
        const off = document.offsetAt(new vscode.Position(line, 0));
-        this.renderProposal(document, state, proposal, { start: off, end: off }, false);
+        this.recordProposal(state, proposal, { start: off, end: off }, false);
      } else {
-        this.renderProposal(document, state, proposal, resolved, true);
+        this.recordProposal(state, proposal, resolved, true);
+        // #70: REBUILD the applied-range bookkeeping from an exact resolve when
+        // the map has no entry (a prior-session apply, or after a close/reopen
+        // clear). An entry already present has been shift-maintained continuously
+        // since apply and is ground truth — never clobbered by a resolve, which
+        // can land on an exact duplicate of the applied text elsewhere.
+        if (state.applied.has(proposal.id) && !state.appliedSpans.has(proposal.id)) {
+          state.appliedSpans.set(proposal.id, resolved);
+        }
      }
    }
-    this.renderDecorations(document, state);
    this.renderStatus(state);
+    this.fireChanged(document);
  }

  /** Shared-watcher entry point (extension.ts): a sidecar changed externally. */
  handleExternalSidecarChange(uri: vscode.Uri): void {
    for (const state of this.docs.values()) {
      if (this.store.sidecarPath(state.docPath) === uri.fsPath) {
-        const doc = vscode.workspace.textDocuments.find((d) => this.docPathOf(d.uri) === state.docPath);
+        const doc = vscode.workspace.textDocuments.find((d) => this.keyOf(d) === state.docPath);
        if (doc) this.renderAll(doc);
      }
    }
  }

  private onDidChange(e: vscode.TextDocumentChangeEvent): void {
-    const state = this.docs.get(this.docPathOf(e.document.uri));
-    if (!state || state.live.size === 0) return;
+    const state = this.docs.get(this.keyOf(e.document));
+    if (!state || (state.live.size === 0 && state.appliedSpans.size === 0)) return;
    for (const change of e.contentChanges) {
      const edit = { start: change.rangeOffset, end: change.rangeOffset + change.rangeLength, newLength: change.text.length };
-      for (const [id, range] of state.live) {
-        const next = shift(range, edit);
-        state.live.set(id, next);
-        const vsThread = state.vsThreads.get(id);
-        if (vsThread && !state.unresolved.has(id)) {
-          vsThread.range = new vscode.Range(e.document.positionAt(next.start), e.document.positionAt(next.end));
-        }
+      for (const [id, range] of state.live) state.live.set(id, shift(range, edit));
+      // #70: an edit fully inside a tracked applied span (or fully outside it)
+      // keeps the span meaningful; one straddling a boundary — e.g. a whole-buffer
+      // external replace — makes the shifted range a clamped guess, so the span
+      // is distrusted (deleted) rather than reverted-to by guess (INV-11).
+      for (const [id, range] of state.appliedSpans) {
+        const shifted = shiftTracked(range, edit);
+        if (shifted === "distrusted") state.appliedSpans.delete(id);
+        else state.appliedSpans.set(id, shifted);
      }
    }
-    // Live shift keeps the UI following; staleness is judged at decision time
-    // (accept re-resolves, INV-11) and at the next renderAll.
-    this.renderDecorations(e.document, state);
  }

  // ---- rendering -----------------------------------------------------------------------

-  private renderProposal(
-    document: vscode.TextDocument,
-    state: DocState,
-    proposal: Proposal,
-    offsets: OffsetRange,
-    pending: boolean,
-  ): void {
-    const fp = state.artifact.anchors[proposal.anchorId]?.fingerprint;
-    const range = new vscode.Range(document.positionAt(offsets.start), document.positionAt(offsets.end));
-    const vsThread = this.controller.createCommentThread(document.uri, range, [
-      {
-        body: new vscode.MarkdownString(proposalBody(fp?.text ?? "", proposal)),
-        mode: vscode.CommentMode.Preview,
-        author: { name: proposal.author.id },
-      },
-    ]);
-    // Proposals are decide-only (INV-12): ✓ accept / ✗ reject in the title
-    // bar. Without this, VS Code renders its default "Reply…" input with no
-    // submit command wired — a dead end. Discussion belongs in a regular
-    // coauthoring thread; proposal discussion trails are deferred (spec §1.7).
-    vsThread.canReply = false;
-    vsThread.label = pending
-      ? "Pending proposal"
-      : "⚠ Stale proposal (target text changed or missing) — accept disabled";
-    vsThread.contextValue = pending ? "pending" : "unresolved";
-    vsThread.collapsibleState = pending
-      ? vscode.CommentThreadCollapsibleState.Expanded
-      : vscode.CommentThreadCollapsibleState.Collapsed;
-    state.vsThreads.set(proposal.id, vsThread);
+  /**
+   * Record a proposal's resolved anchor in the live/unresolved bookkeeping.
+   * No editor UI (F10/INV-32: the rendered preview is the single review
+   * surface) — this keeps state.live/state.unresolved populated so the preview
+   * (SLICE-3) and the getRendered test seam can read it.
+   */
+  private recordProposal(state: DocState, proposal: Proposal, offsets: OffsetRange, pending: boolean): void {
    state.live.set(proposal.id, offsets);
    if (!pending) state.unresolved.add(proposal.id);
  }

-  private renderDecorations(document: vscode.TextDocument, state: DocState): void {
-    const ranges: vscode.Range[] = [];
-    for (const [id, off] of state.live) {
-      if (state.unresolved.has(id)) continue;
-      ranges.push(new vscode.Range(document.positionAt(off.start), document.positionAt(off.end)));
-    }
-    for (const editor of vscode.window.visibleTextEditors) {
-      if (editor.document === document) editor.setDecorations(this.pendingType, ranges);
-    }
-  }
-
  private renderStatus(state: DocState): void {
    const n = state.unresolved.size;
    if (n === 0) {
@@ -297,18 +646,7 @@ export class ProposalController implements vscode.Disposable {
  // ---- lookups ----------------------------------------------------------------------------

  private openDoc(state: DocState): vscode.TextDocument | undefined {
-    return vscode.workspace.textDocuments.find((d) => this.docPathOf(d.uri) === state.docPath);
-  }
-  private byThread(vsThread: vscode.CommentThread): { state: DocState; proposal: Proposal } | undefined {
-    for (const state of this.docs.values()) {
-      for (const [id, t] of state.vsThreads) {
-        if (t === vsThread) {
-          const proposal = state.artifact.proposals.find((p) => p.id === id);
-          if (proposal) return { state, proposal };
-        }
-      }
-    }
-    return undefined;
+    return vscode.workspace.textDocuments.find((d) => this.keyOf(d) === state.docPath);
  }
  private byId(docPath: string, proposalId: string): { state: DocState; proposal: Proposal } | undefined {
    const state = this.docs.get(docPath);
@@ -322,16 +660,9 @@ export class ProposalController implements vscode.Disposable {
    const state = this.docs.get(docPath);
    if (!state) return [];
    const out: RenderedProposal[] = [];
-    for (const [id, vsThread] of state.vsThreads) {
+    for (const [id, off] of state.live) {
      const p = state.artifact.proposals.find((x) => x.id === id)!;
-      const off = state.live.get(id)!;
-      out.push({
-        id,
-        pending: !state.unresolved.has(id),
-        canReply: vsThread.canReply !== false,
-        turnId: p.turnId,
-        range: { start: off.start, end: off.end },
-      });
+      out.push({ id, pending: !state.unresolved.has(id), canReply: false, turnId: p.turnId, range: { start: off.start, end: off.end } });
    }
    return out;
  }
@@ -12,7 +12,7 @@ export function addProposal(
  fp: Fingerprint,
  replacement: string,
  author: Provenance,
-  opts?: { turnId?: string; instruction?: string },
+  opts?: { turnId?: string; instruction?: string; granularity?: "block" | "single" },
 ): { proposalId: string; anchorId: string } {
  const anchorId = newId("a");
  const proposalId = newId("pr");
@@ -25,6 +25,7 @@ export function addProposal(
    createdAt: new Date().toISOString(),
    ...(opts?.turnId !== undefined ? { turnId: opts.turnId } : {}),
    ...(opts?.instruction !== undefined ? { instruction: opts.instruction } : {}),
+    ...(opts?.granularity !== undefined ? { granularity: opts.granularity } : {}),
  });
  return { proposalId, anchorId };
 }
@@ -36,6 +37,25 @@ export function removeProposal(artifact: Artifact, proposalId: string): boolean
  return artifact.proposals.length < before;
 }

+/**
+ * F12/#64 (INV-48): record a proposal as optimistically applied — store the
+ * pre-apply `original` (for revert + the struck `<del>`) and re-point its anchor
+ * fingerprint to the now-in-buffer applied text so `resolve()` finds it. Idempotent
+ * shape: a second call simply overwrites with the same values.
+ */
+export function setProposalApplied(
+  artifact: Artifact,
+  proposalId: string,
+  appliedFp: Fingerprint,
+  original: string,
+): boolean {
+  const p = artifact.proposals.find((x) => x.id === proposalId);
+  if (!p) return false;
+  p.original = original;
+  artifact.anchors[p.anchorId] = { fingerprint: appliedFp };
+  return true;
+}
+
 /** Markdown comment body: instruction header + fenced whole-range diff. */
 export function proposalBody(targetText: string, p: Proposal): string {
  const header = p.instruction ? `**Claude proposes** — _${p.instruction}_` : "**Claude proposes**";
@@ -0,0 +1,120 @@
+/**
+ * ScmSurfaceController — the native diff surface (spec §6.4/§5, INV-13).
+ * Owns: the "Cowriting" SourceControl + QuickDiffProvider (gutter bars), the
+ * cowriting-baseline: TextDocumentContentProvider, the "Review Changes" native
+ * diff command, and the "✦ Coediting · N changes" status-bar item. Gated on
+ * the CoeditingRegistry (INV-10): non-entered docs resolve no original.
+ */
+import * as path from "node:path";
+import * as vscode from "vscode";
+import type { CoeditingRegistry } from "./coeditingRegistry";
+import type { DiffViewController } from "./diffViewController";
+import { countLineHunks } from "./trackChangesModel";
+
+export const BASELINE_SCHEME = "cowriting-baseline";
+
+export function baselineUriFor(docUri: vscode.Uri): vscode.Uri {
+  const name = path.basename(docUri.path) || "untitled.md";
+  return vscode.Uri.from({ scheme: BASELINE_SCHEME, path: `/${name}`, query: encodeURIComponent(docUri.toString()) });
+}
+
+export class ScmSurfaceController implements vscode.Disposable {
+  private readonly disposables: vscode.Disposable[] = [];
+  private readonly statusItem: vscode.StatusBarItem;
+  private readonly counts = new Map<string, number>();
+  private readonly baselineEmitter = new vscode.EventEmitter<vscode.Uri>();
+  private readonly recountTimers = new Map<string, ReturnType<typeof setTimeout>>();
+
+  constructor(
+    private readonly registry: CoeditingRegistry,
+    private readonly diffView: DiffViewController,
+  ) {
+    this.statusItem = vscode.window.createStatusBarItem(vscode.StatusBarAlignment.Right, 100);
+    this.statusItem.command = "cowriting.reviewChanges";
+    this.statusItem.tooltip = "Cowriting — click to review changes (baseline ⟷ live)";
+    this.disposables.push(this.statusItem, this.baselineEmitter);
+
+    this.disposables.push(
+      vscode.workspace.registerTextDocumentContentProvider(BASELINE_SCHEME, {
+        onDidChange: this.baselineEmitter.event,
+        provideTextDocumentContent: (uri) =>
+          this.diffView.getBaseline(decodeURIComponent(uri.query))?.text ?? "",
+      }),
+    );
+
+    const sc = vscode.scm.createSourceControl("cowriting", "✦ Cowriting");
+    sc.quickDiffProvider = {
+      provideOriginalResource: (uri: vscode.Uri) =>
+        this.registry.isCoediting(uri) && this.diffView.getBaseline(uri.toString())
+          ? baselineUriFor(uri)
+          : null,
+    };
+    this.disposables.push(sc);
+
+    this.disposables.push(
+      vscode.commands.registerCommand("cowriting.reviewChanges", () => this.openReview()),
+      this.diffView.onDidChangeBaseline(({ uri }) => {
+        this.baselineEmitter.fire(baselineUriFor(vscode.Uri.parse(uri)));
+        this.scheduleRecount(uri);
+      }),
+      this.registry.onDidChange(({ uri, coediting }) => {
+        if (coediting) this.scheduleRecount(uri);
+        this.refreshStatus();
+      }),
+      vscode.workspace.onDidChangeTextDocument((e) => {
+        if (this.registry.isCoediting(e.document.uri)) this.scheduleRecount(e.document.uri.toString());
+      }),
+      vscode.window.onDidChangeActiveTextEditor(() => this.refreshStatus()),
+    );
+    this.refreshStatus();
+  }
+
+  changeCount(uriString: string): number {
+    return this.counts.get(uriString) ?? 0;
+  }
+
+  private async openReview(): Promise<void> {
+    const ed = vscode.window.activeTextEditor;
+    if (!ed || !this.registry.isCoediting(ed.document.uri)) {
+      void vscode.window.showWarningMessage("Cowriting: not coediting this document.");
+      return;
+    }
+    this.baselineEmitter.fire(baselineUriFor(ed.document.uri));
+    await vscode.commands.executeCommand(
+      "vscode.diff",
+      baselineUriFor(ed.document.uri),
+      ed.document.uri,
+      `${path.basename(ed.document.uri.path)} — Coediting (baseline ⟷ live)`,
+      { preview: true },
+    );
+  }
+
+  private scheduleRecount(uriString: string): void {
+    const prev = this.recountTimers.get(uriString);
+    if (prev !== undefined) clearTimeout(prev);
+    this.recountTimers.set(uriString, setTimeout(() => {
+      this.recountTimers.delete(uriString);
+      const doc = vscode.workspace.textDocuments.find((d) => d.uri.toString() === uriString);
+      const baseline = this.diffView.getBaseline(uriString);
+      if (!doc || !baseline) return;
+      this.counts.set(uriString, countLineHunks(baseline.text, doc.getText()));
+      this.refreshStatus();
+    }, 150));
+  }
+
+  private refreshStatus(): void {
+    const ed = vscode.window.activeTextEditor;
+    if (!ed || !this.registry.isCoediting(ed.document.uri)) {
+      this.statusItem.hide();
+      return;
+    }
+    const n = this.counts.get(ed.document.uri.toString()) ?? 0;
+    this.statusItem.text = `$(sparkle) Coediting · ${n} change${n === 1 ? "" : "s"}`;
+    this.statusItem.show();
+  }
+
+  dispose(): void {
+    for (const t of this.recountTimers.values()) clearTimeout(t);
+    for (const d of this.disposables) d.dispose();
+  }
+}
--- a/Show More
+++ b/Show More