Replaces the jargon 'Pin Review Baseline to Now' (snapshot-mode reset) with
the user-intent label 'Mark Changes as Reviewed', which fits the review
framing of the product. Label-only; semantics unchanged (snapshot-mode only).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- D14: 'Pin Review Baseline to Now' exists only in snapshot mode (no committed
HEAD); git-tracked files have no manual pin -> commit is the reset; pin
command when-gated on cowriting.baselineMode==snapshot. Removed open Q8.
- INV-8: changes already present at coediting start (uncommitted vs HEAD,
incl. edits made outside this experience) are attributed to the writer
(human) -- session can't know their origin; surfaces prior work as human.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Answers 'what about docs not in a git repo' (~/Downloads, untitled): a
BaselineRouter mirroring the SidecarRouter. Git-tracked file -> baseline =
HEAD blob (read via Git extension API, auto-resets on commit, storage-free);
no commit to anchor to -> snapshot (set on entry, re-pinnable, global
storage/in-memory). Manual pin works in both (precedes HEAD until next
commit); new in-repo file flips snapshot->HEAD on first commit. Updated
INV-7, data model, baseline router, PUC-1/7, decisions, deps; new open Q8
(pin-ahead-of-HEAD precedence in git files).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Entering coediting sets the baseline to current content automatically;
'Pin Review Baseline to Now' resets it ('everything so far is settled; show
changes from here'). A coedited doc always has a baseline -> removed the
'no/missing baseline' empty state entirely; the only empty state is a
genuinely empty document. Updated INV-7, PUC-1/7, baseline provider, preview
(marks since checkpoint), failure modes, flags, glossary.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Authorship lives in the annotated preview (D11); native diff stays
author-blind. Crucially, the review's proposed side is WRITABLE (INV-12,
D12): the writer tweaks the machine's suggestion before Keep ('close, but a
few human edits'), the live doc is untouched until Keep (INV-5), and on Keep
machine words vs writer tweaks are attributed separately by diffing
final<->machine-original<->baseline. This keeps authorship clean AT CREATION,
so post-hoc in-editor authorship (opts 2/3) is unnecessary and deferred. New
open Q7: per-change Keep/Undo semantics after free cross-hunk editing.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- INV-10: coediting is opt-in per document ('Coedit this Document with
Claude'); every surface (SCM/QuickDiff, comments, preview annotations,
edit entry points) gates on the active coediting set so other markdown
files are never hijacked. Adds PUC-7 (enter/leave), CoeditingRegistry.
- BUC-5/PUC-8/D8: on a coedited doc a comment fires a turn that replies
in-thread and offers an edit -> pending proposal (never edits directly).
- INV-11/D9: verified against the VS Code API that the native diff editor
cannot recolor changes per-author (diffEditor.* tokens are global;
inlineEdit./chat tokens are accept-state not author; diff-pane decorations
composite + are unreliable). Authorship lives in the preview (optionally a
decoration track-changes editor). New open Qs: Q5 (authorship surface),
Q6 (comment trigger).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Reframe §1 away from a redesign-of-the-old-plugin narrative and root it in
the real-world problem: coediting a markdown document with a machine is hard
today (rewrite-not-a-diff, all-or-nothing, blurred authorship, no clean
rendered view, opaque machine work). Solution and technical substance stand on
their own; proven patterns reused on merit, not as old-plugin lineage.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
BenStullsBets