v0.30.2: header nav label "Philosophy" -> "About"

Revert the §14.3 persistent chrome link's display text from "Philosophy" back to "About" (relabeled in v0.21.0). Display text only — route /philosophy, the header-about class, and the page are unchanged. Patch per SPEC §20.2: cosmetic, no deployment action beyond a frontend rebuild. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
v0.30.1: fix phantom pending-idea after merge-with-branch-delete
2026-05-29 06:56:14 -07:00 · 2026-05-29 05:26:00 -07:00 · 2026-05-29 02:12:18 -07:00 · 2026-05-29 03:11:02 +00:00 · 2026-05-28 20:10:13 -07:00 · 2026-05-29 02:42:05 +00:00
42 changed files with 2507 additions and 243 deletions
@@ -23,6 +23,258 @@ skip versions are the composition of each intervening adjacent
 release's steps in order — no A-to-B path is pre-computed beyond
 that.

+## 0.30.2 — 2026-05-29
+
+**Patch — header nav label: the persistent chrome link reverts from
+"Philosophy" back to "About." Display text only — the route
+(`/philosophy`), the `header-about` class, and the page itself are
+unchanged. A plain frontend rebuild applies it; no schema, API, config,
+overlay, or secret change.**
+
+The §14.3 persistent link was relabeled "About" → "Philosophy" in
+v0.21.0. This restores "About" as the neutral, framework-native label
+(the CSS class `header-about` and the surrounding comment already call
+it "the About link"). A deployment that wants a more pointed framing
+can title its own About page in the `PHILOSOPHY.md` content the
+`PHILOSOPHY_PATH` override serves.
+
+Upgrade steps: none. **SHOULD** deploy as a normal code deploy.
+
+## 0.30.1 — 2026-05-29
+
+**Patch — bug fix: a merged idea-PR whose branch was deleted no longer
+lingers as a phantom "pending idea." No schema, API, config, overlay,
+or secret change — a plain code deploy applies it, and the running
+reconciler clears any existing ghost on the next sweep (≤5 min) once
+deployed. Shipped from driver session 0040.0.**
+
+`refresh_meta_pulls` (and `refresh_rfc_repo`) recover a PR's slug/kind
+by parsing its Gitea `head.ref`. When a PR is merged **and its branch
+deleted**, Gitea stops reporting the real branch name and returns the
+synthetic `refs/pull/<N>/head` sentinel instead. The slug then parsed
+to `None`, the reconcile loop skipped the row, and `cached_prs.state`
+stayed frozen at `open` forever — so the entry showed as **both** a
+super-draft (the `cached_rfcs` push-event reconcile succeeded) **and** a
+pending idea (the `cached_prs` PR-close reconcile never landed). The fix
+recovers the original branch name from the already-stored `cached_prs`
+row (which retains the real `head_branch` from when the PR was open;
+migration 002) whenever Gitea reports an empty or `refs/pull/` sentinel
+ref. Regression test added in `test_propose_vertical.py`
+(`test_merged_idea_pr_with_deleted_branch_clears_proposal`).
+
+Surfaced through the ROADMAP #35 operator authoring lane, which merges
+idea PRs from the CLI with branch-deletion enabled — a path the web UX
+never exercises (it leaves branches in place, so
+`default_delete_branch_after_merge` stays false). The framework should
+not depend on branches outliving their merge, hence the framework-level
+fix rather than a tooling workaround.
+
+Upgrade steps: none. **SHOULD** deploy as a normal code deploy; the
+periodic reconciler self-heals any existing phantom on its next sweep.
+
+## 0.30.0 — 2026-05-29
+
+**Minor — documentation: the user guide (`DOCS.md`, served at
+`/docs/user-guide` via `/api/docs`) brought back in sync with the
+shipped app. No code, schema, API, config, overlay, or secret change —
+a plain code deploy serves the updated guide. Shipped from driver
+session 0037.0.**
+
+The guide had drifted since it was first written: it still described
+the pre-OTC email *allowlist* sign-in, listed four propose-RFC fields,
+and predated several shipped surfaces. Updated to match v0.7.0–v0.29.0:
+
+- **Signing in** rewritten for the email + one-time-code flow (v0.7.0),
+  optional passcode (v0.10.0), trust-this-device for 30 days (v0.11.0),
+  optional Cloudflare Turnstile (v0.12.0), and the beta-access request →
+  `pending` → admin-`granted` gate (v0.8.0 / #6), plus the admin-create
+  + invite-claim path (v0.17.0 / #16). The vestigial allowlist is no
+  longer described as the gate.
+- **Proposing a new RFC** now lists five fields, adding the optional
+  "What will you be using this RFC for?" use-case field (#26) and noting
+  the AI tag-suggestion disclosure (#27).
+- **Roles & permissions** documents the `pending` state.
+- New **Invitations, cross-references, and contribution requests**
+  section covers owner invitations (#12) and the RFC auto-link /
+  create-RFC / ask-to-contribute affordances (#28).
+- New **Privacy and cookies** section covers the consent banner and
+  consent-gated analytics (#11 / #13).
+
+Upgrade steps: none — documentation-only; the change is the `DOCS.md`
+file served verbatim by `/api/docs`. A plain code deploy at this tag
+serves it. A deployment that overrides the guide via `DOCS_PATH`
+supplies its own copy and is unaffected.
+
+## 0.29.0 — 2026-05-28
+
+**Minor — roadmap #28 Parts 2 + 3: offer-to-create-an-RFC for strong-
+candidate terms, and offer-to-contribute-to-a-pending-RFC. One auto-
+applied migration (024, additive: a new `contribution_requests` table).
+No config/overlay/secret change; no nginx/systemd change. A plain code
+deploy + the auto-migration picks it up. Shipped from driver session
+0033.0.**
+
+Both parts extend the v0.26.0 (#28 Part 1) read-time scanner
+(`backend/app/rfc_links.py`) and its renderer
+(`frontend/src/components/LinkedText.jsx`). The scanner now sorts each
+matched term into one of three buckets — active link (Part 1, unchanged),
+pending-RFC contribute offer (Part 3), create-RFC offer (Part 2) — in one
+pass, with precedence active > pending > candidate at any position. The
+backend still emits only structured segments (never HTML), so the surface
+stays XSS-safe by construction.
+
+- **Part 2 — create-RFC offers.** A *strong-candidate* term — a
+  **multi-word tag** from the #27 tag taxonomy that has no defining RFC
+  (no active or super-draft RFC whose slug/title is that term) — renders,
+  for a viewer with create rights (`permission_state='granted'`), as an
+  inline "+ create RFC" affordance. Clicking opens the propose-RFC modal
+  with the term pre-filled as the title (`ProposeModal` gained an
+  `initialTitle`; the affordance routes via `?propose=<term>`, read in
+  `App.jsx`). The heuristic is deliberately conservative — multi-word is
+  the same false-positive guard the title rule uses, so a single common
+  tag word (`identity`) is never offered. Broader candidate detection
+  (capitalized phrases mined from text, terms repeated across recent PRs,
+  or the #27 Haiku `ANTHROPIC_API_KEY` pathway) is a sanctioned but
+  deferred extension.
+- **Part 3 — contribute-to-pending offers.** A term matching a *pending*
+  RFC — a super-draft (`state='super-draft'`: accepted as an idea, owned,
+  with a contribution surface, not yet graduated) — renders, for a
+  signed-in non-owner, as an inline "ask to contribute" affordance
+  carrying the owner's display name ("<owner> is working on an RFC for
+  '<term>'"). It opens a contribute-request form (`?contribute=<slug>`)
+  with three fields — **who I am** (required), **why I'm asking**
+  (required), **what I'd use it for** (optional, mirroring #26). Submitting
+  lands a `contribution_requests` row and one actionable §15 inbox
+  notification per owner (new kind `contribution_request_on_pending_rfc`,
+  category `personal-direct` — so it reuses the existing
+  `email_personal_direct` preference, no new toggle). In the inbox the
+  owner sees the requester's who/why/use-case inline with **Accept** /
+  **Decline**. Accept fires #12's owner-invite flow with the requester as
+  the invitee (a `contributor` `rfc_invitations` row + the existing invite
+  email) and echoes a notification back to the requester; Decline closes
+  the request and notifies the requester. Pre-merge idea PRs (not yet in
+  `cached_rfcs`, no contribution surface) are deliberately out of scope —
+  a documented future extension.
+
+New endpoints (all under the existing `/api` router):
+`GET /api/rfcs/{slug}/contribution-target`,
+`POST /api/rfcs/{slug}/contribution-requests`,
+`POST /api/rfcs/{slug}/contribution-requests/{id}/accept`,
+`POST /api/rfcs/{slug}/contribution-requests/{id}/decline`.
+
+The owner-invite issue path was refactored into one reusable chokepoint,
+`api_invitations.issue_invitation(...)`, shared by the manual invite
+endpoint and Part 3's accept path so the dup-guard, token mint, insert,
+and transactional email stay identical.
+
+Upgrade steps:
+
+1. Deployments **MUST** apply the auto-run migration `024` (additive: the
+   new `contribution_requests` table; no existing table or row is
+   touched). The standard deploy path runs pending migrations on start —
+   no manual step beyond deploying the new code.
+2. No config, overlay, or secret change is required. The Part 2 candidate
+   affordance reuses #27's tag taxonomy; it surfaces only when the corpus
+   carries multi-word tags without a defining RFC, and the create
+   affordance renders only for beta-granted viewers. Part 3's email reuse
+   sends through the existing invitation SMTP path — no new key.
+
+## 0.28.0 — 2026-05-28
+
+**Minor — security-hardening follow-up (Session-0026 audit, informational
+findings I3 + I4). No operator action required: no migration, no schema
+change, no config/overlay change, no API/behavior change for any caller.
+A plain code deploy picks it up. Shipped from driver session 0032.0.**
+
+Two informational findings from the Session-0026 audit, both
+framework-internal defense-in-depth:
+
+- **I3 — dead HTML-email branch guarded.** `email_envelope.build_envelope`
+  accepted a `body_html=` argument that built a `multipart/alternative`
+  body, but no send path ever passed it — every rfc-app mail is plain
+  text. An unused branch that would emit HTML built from (potentially
+  unescaped) user content is the C1 stored-XSS class waiting in the mail
+  channel. The branch is now a loud guard: passing `body_html` raises
+  `NotImplementedError`. The argument is kept in the signature for
+  documented future symmetry; enabling HTML mail becomes a deliberate
+  change that MUST HTML-escape user content at the call site and remove
+  the guard in the same commit.
+- **I4 — Turnstile siteverify no longer blocks the event loop.**
+  `turnstile.verify_token` was a synchronous function issuing a blocking
+  `httpx.post` from inside the async `/auth/otc/request` handler, so a
+  slow CloudFlare response stalled the single worker for up to the 10s
+  timeout. It is now `async` and awaits the call on an
+  `httpx.AsyncClient` (matching the codebase's existing async-httpx
+  pattern), isolated behind a narrow `_siteverify_post` seam. The sole
+  caller (`main.py`) now `await`s it.
+
+Upgrade steps: **none.** Both changes are internal. The `verify_token`
+signature changed from sync to `async` (callers must `await`), but the
+only caller is in-tree (`main.py`) and is updated in this release; no
+deployment-facing surface, config key, or migration is affected.
+
+## 0.27.0 — 2026-05-28
+
+**Minor — security-hardening release (Session-0026 audit remediation).
+One auto-applied migration (023); one behavior change that re-prompts
+device-trust; deployments MUST re-apply the nginx + systemd files.**
+This is the work cut as the "v0.25.0 security-hardening" branch; it
+reversioned to 0.27.0 because v0.26.0 (#28) took the next slot while it
+was in flight. Shipped from driver session 0030.0.
+
+- **C1 (Critical) — stored-XSS closed.** Every markdown→HTML sink now
+  routes through one chokepoint, `frontend/src/lib/sanitizeHtml.js`
+  (DOMPurify), before any `innerHTML` / `dangerouslySetInnerHTML` write:
+  `MarkdownPreview`, both `ProposalView` sinks (entry body +
+  `proposed_use_case`), and `Editor`. A hook adds
+  `rel="noopener noreferrer"` to `target=_blank` links. `marked` no
+  longer passes raw HTML / `javascript:` URIs to the DOM, so a
+  contributor can no longer plant a payload that runs in an admin/owner
+  session during review.
+- **H1 — OTC verify is rate-limited.** New `backend/app/ratelimit.py`
+  (per-IP token buckets) gates `/auth/otc/verify`, `/auth/otc/request`,
+  and the passcode check/verify paths; a per-account OTC-verify lockout
+  (migration `023_otc_verify_lockout.sql`) mirrors the passcode lockout.
+- **M1 — device-trust lookup no longer table-scans.** The device-trust
+  cookie value is now `"<row_id>.<raw_token>"`; `device_trust.lookup`
+  reads the one indexed row and bcrypt-checks only it, instead of
+  bcrypt-checking every row in the table on each unauthenticated
+  `/auth/device-trust/start`.
+- **M2 — HTTP security headers** (CSP, HSTS, X-Frame-Options,
+  X-Content-Type-Options, Referrer-Policy) added to the nginx server
+  block. **L8/I1**: `server_tokens off` + legacy TLS1.0/1.1 removed.
+- **M4 — session cookie `Secure` by default** (`SESSION_COOKIE_SECURE`,
+  defaults on; a dev box on plain http sets it `false`).
+- **M5 — bounce webhook fails closed.** An unset
+  `WEBHOOK_EMAIL_BOUNCE_SECRET` now **disables** `/api/webhooks/email-bounce`
+  (503) instead of leaving it open; a dev opts back in with
+  `RFC_APP_INSECURE_BOUNCE_WEBHOOK=1`.
+- **L2/L3** per-IP cooldown + check-endpoint throttle. **L4** systemd
+  sandbox knobs (`CapabilityBoundingSet=`, `ProtectKernel*`,
+  `RestrictAddressFamilies`, `SystemCallFilter`, …).
+
+Upgrade steps:
+
+1. **Migration** — none manual; `023_otc_verify_lockout.sql` auto-applies
+   at startup via `db.run_migrations`.
+2. **Device trust (MUST expect re-prompt)** — the cookie format changed,
+   so existing "trusted device" cookies no longer match; affected users
+   are re-prompted for device verification once. No data migration; old
+   rows are simply never matched and age out.
+3. **nginx + systemd (MUST apply out-of-band)** — the deploy gesture does
+   **not** install `deploy/nginx/ohm.wiggleverse.org.conf` or
+   `deploy/systemd/rfc-app.service`. After deploying the code, copy both
+   to their system locations, then `nginx -t && systemctl reload nginx`
+   and `systemctl daemon-reload && systemctl restart <unit>`. (M2 headers
+   and L4 sandboxing do not take effect until this is done.)
+4. **Bounce webhook (SHOULD)** — bind `WEBHOOK_EMAIL_BOUNCE_SECRET` (or
+   set `RFC_APP_INSECURE_BOUNCE_WEBHOOK=1` for dev). Unset → the endpoint
+   returns 503 (closed). No legitimate bounce source is wired today, so
+   503 is the safe default.
+5. **Session cookie (SHOULD, dev only)** — a deployment served over plain
+   http MUST set `SESSION_COOKIE_SECURE=false` or the session cookie
+   won't be sent. Production over HTTPS leaves it unset (Secure on).
+
 ## 0.26.0 — 2026-05-28

 **Minor — no schema migration, no new secret, no config, no upgrade
@@ -39,23 +39,44 @@ thread. Every write affordance is replaced with a sign-in prompt.

 ## Signing in

-While the framework is in private beta, only invited email addresses
-can complete sign-in. If your email is on the allowlist, the
-"Sign in" button in the header completes the flow and lands you on
-the catalog with full read and write access. If your email is not on
-the allowlist, you'll be sent to a short "pending" page explaining
-the gate.
+Anyone can start the sign-in flow with their own email address — there
+is no invite-only allowlist. Sign-in is passwordless:

-Once you have an account, you're a **contributor** by default — the
-role that grants every write affordance the app exposes, scoped by
-the per-RFC and per-branch rules described below.
+1. **Enter your email.** If the deployment has human verification
+   enabled (a Cloudflare Turnstile challenge), you complete it here.
+2. **Enter the one-time code.** The app emails you a short numeric
+   code; entering it signs you in. Codes expire after a few minutes,
+   and repeated wrong entries briefly lock the email.
+3. **Set a passcode (optional).** After your first code sign-in you
+   can set a passcode. On later visits you sign in with email +
+   passcode, with the one-time code as the forgot-passcode fallback.
+4. **Trust this device (optional).** You can mark a device trusted for
+   30 days to skip the code/passcode step on it. Trusted devices are
+   listed in your settings and can be revoked individually or all at
+   once.
+
+### Getting write access
+
+Signing in gives you an account, but write access is gated. The first
+time you sign in you're asked for your first name, last name, and a
+short note on why you'd like access; you then land on a "request in
+review" page. While your account is **pending**, you can read
+everything an anonymous visitor can but cannot write — no chat,
+propose, branch, PR, or discussion post. Once an admin **grants** your
+account you become a **contributor**, the role that carries every
+write affordance the app exposes, scoped by the per-RFC and per-branch
+rules described below.
+
+An admin can also create your account ahead of time and email you an
+invite link. Clicking it claims the account and signs you in with the
+role the admin assigned, skipping the one-time-code step.

 ---

 ## Proposing a new RFC

 A new RFC begins as a proposal. The "+ Propose new RFC" button at
-the bottom of the catalog opens a small modal that collects four
+the bottom of the catalog opens a small modal that collects five
 things:

 - **Title.** The word, concept, or topic this RFC would define.
@@ -65,8 +86,13 @@ things:
  inline.
 - **Pitch.** One or two paragraphs answering *why this RFC is
  needed*. This becomes the body of the entry.
- **Tags.** Optional. The AI suggests tags from the pitch; you can
-  accept, dismiss, or type your own.
+- **Use case.** Optional. *What will you be using this RFC for?* —
+  the concrete application driving the proposal, as distinct from the
+  abstract case for it. Leaving it blank is fine.
+- **Tags.** Optional. If the deployment has AI tag suggestion
+  enabled, suggested tags appear as you fill the form (with an inline
+  note that the text you've entered is sent to the model that
+  generates them); you can accept, dismiss, or type your own.

 Submitting the modal does one concrete thing: it opens a pull
 request against the framework's meta repository, adding one new
@@ -167,6 +193,51 @@ either requires a contributor account.

 ---

+## Invitations, cross-references, and contribution requests
+
+Three connected surfaces help the right people find and join the
+right RFC.
+
+### Owner invitations
+
+An RFC's owner (or an app-wide admin or owner) can invite a specific
+person to that RFC from the "Invitations" control in the RFC header.
+The invite names an email and a role for *this RFC*:
+
+- **contributor** — can open PRs and join the discussion;
+- **discussant** — can join the discussion only.
+
+The invitee gets an email with an accept link; accepting adds them as
+a collaborator on that RFC. The invitations panel lists every invite
+with its status (pending / accepted / expired / revoked); pending
+invites can be revoked. An invitation is per-RFC — it does not change
+the invitee's app-wide role, and it cannot lift the pending gate: the
+invitee still needs a granted account to write.
+
+### RFC cross-links in PRs and comments
+
+When a PR description or a comment mentions an existing active RFC —
+by its ID, its multi-word title, or its slug — the framework renders
+that mention as a link to the RFC. The matching is conservative by
+design (single common words are never auto-linked), and the links are
+computed at read time, so nothing is rewritten in what you typed.
+
+### "Create" and "ask to contribute" offers
+
+The same scan surfaces two affordances inline:
+
+- If a term looks like it should have an RFC but none exists yet, a
+  reader who has create rights sees a **"create RFC for '<term>'"**
+  link that opens the propose modal with the title pre-filled.
+- If a term matches a *pending* RFC (a super-draft someone already
+  owns), a signed-in reader sees an **"ask to contribute"** offer
+  naming the owner. It opens a short request form — who you are, why
+  you're asking, and optionally what you'd use the RFC for. The
+  request lands in the owner's inbox; the owner can **accept** (which
+  sends you an owner invitation) or **decline** (which notifies you).
+
+---
+
 ## Working on a branch

 Contribute mode flips one branch into edit-enabled. The centre
@@ -505,6 +576,14 @@ Each role is a strict superset of the one below it.
   entirely. The framework names a single "owner zero" at
   bootstrap.

+Between anonymous and contributor sits one transient state:
+**pending**. A freshly signed-in account that hasn't been granted
+access yet (see [Signing in](#signing-in)) reads everything an
+anonymous visitor can, but no write affordance unlocks until an admin
+grants it. Granting promotes the account to contributor; an admin can
+also revoke a granted account back to a no-write state. These
+transitions are recorded in the `permission_events` log.
+
 The practical difference between admin and owner is narrow but
 load-bearing: admin is the operational tier — it does the day-to-
 day moderation and stewardship work; owner is the tier that
@@ -594,6 +673,20 @@ review.

 ---

+## Privacy and cookies
+
+A consent banner appears on your first visit and lets you choose
+which cookie categories to allow — essential always, with analytics
+and other categories opt-in. The choice is remembered and can be
+changed any time from the privacy/cookies controls in settings.
+
+Analytics only load if you opt in: the framework defers the analytics
+SDK behind your consent, so declining means it is never initialized.
+The `/privacy` and `/cookies` pages describe what's collected and
+why; a deployment can point those pages at its own fuller policy.
+
+---
+
 ## Where to learn more

 - The framework's *why* lives in [the philosophy
@@ -1 +1 @@
-0.26.0
+0.30.2
@@ -21,6 +21,7 @@ from pydantic import BaseModel, Field
 from . import (
    api_admin,
    api_branches,
+    api_contributions,
    api_discussion,
    api_graduation,
    api_invitations,
@@ -141,6 +142,10 @@ def make_router(
    # invited users keep read access but cannot write (v0.6.0
    # contract extended to per-RFC scope).
    router.include_router(api_invitations.make_router())
+    # v0.29.0 (roadmap item #28 Part 3): offer-to-contribute-to-a-pending
+    # (super-draft) RFC. Reuses the #12 invite flow (api_invitations above)
+    # on accept; lands the request + owner notifications via §15 notify.
+    router.include_router(api_contributions.make_router())

    # ---------------------------------------------------------------
    # §17: /api/health — unauthenticated post-flight probe.
@@ -0,0 +1,311 @@
+"""v0.29.0 / roadmap #28 Part 3 — offer-to-contribute-to-a-pending-RFC.
+
+When the #28 scanner (see ``rfc_links.py``) matches a term in submitted
+PR/comment text to a **pending** RFC — a super-draft
+(``cached_rfcs.state='super-draft'``: accepted as an idea, owned, with a
+contribution surface, but not yet graduated to an active RFC) — the
+reader is offered an "ask to contribute" popover. This module is the
+backend for that flow:
+
+  * ``GET  /api/rfcs/{slug}/contribution-target``      — what the
+    contribute form needs (RFC title, owner display, the viewer's
+    eligibility + whether they already have a pending ask).
+  * ``POST /api/rfcs/{slug}/contribution-requests``    — submit the ask
+    (who-I-am / why / optional use-case); lands a row + one §15
+    notification per owner.
+  * ``POST /api/rfcs/{slug}/contribution-requests/{id}/accept``  — owner:
+    accept, which fires #12's owner-invite flow with the requester as the
+    invitee (opening the RFC's discussion/contribution surface), then
+    echoes a notification back to the requester.
+  * ``POST /api/rfcs/{slug}/contribution-requests/{id}/decline`` — owner:
+    decline; the request closes and the requester is notified.
+
+"Pending" is scoped to a super-draft because that is the state with an
+owner to route to, a contribution surface to open, and a row in
+``cached_rfcs`` for the ``rfc_invitations`` FK the accept path reuses.
+Pre-merge idea PRs are deliberately out of scope (no contribution
+surface yet) — a documented future extension, mirroring the
+conservative scoping in ``rfc_links.py``.
+"""
+from __future__ import annotations
+
+import sqlite3
+from typing import Any
+
+from fastapi import APIRouter, HTTPException, Request
+from pydantic import BaseModel, Field
+
+from . import api_invitations, auth, db, notify
+
+# Field caps — generous for free text, bounded so a request row (and the
+# notification payload that carries it) can't be used to store unbounded
+# blobs. Mirrors the order-of-magnitude of the propose/tag-suggest caps.
+_WHO_MAX = 2000
+_WHY_MAX = 4000
+_USE_CASE_MAX = 4000
+_TERM_MAX = 200
+
+
+class ContributionRequestBody(BaseModel):
+    # The term in the PR/comment text that surfaced the offer (the
+    # super-draft's title/slug). Carried for the owner's context line.
+    matched_term: str = Field(min_length=1, max_length=_TERM_MAX)
+    who_i_am: str = Field(min_length=1, max_length=_WHO_MAX)
+    why: str = Field(min_length=1, max_length=_WHY_MAX)
+    use_case: str | None = Field(default=None, max_length=_USE_CASE_MAX)
+
+
+def _require_super_draft(slug: str):
+    """The contribute surface only operates on a *pending* RFC. 404 on
+    unknown; 409 on a state that isn't a super-draft (active RFCs use the
+    Part-1 link, not a contribute offer; withdrawn is closed)."""
+    row = db.conn().execute(
+        "SELECT slug, title, state, owners_json, proposed_by FROM cached_rfcs WHERE slug = ?",
+        (slug,),
+    ).fetchone()
+    if row is None:
+        raise HTTPException(404, "RFC not found")
+    if row["state"] != "super-draft":
+        raise HTTPException(409, "RFC is not a pending super-draft")
+    return row
+
+
+def _require_request(slug: str, request_id: int):
+    row = db.conn().execute(
+        """
+        SELECT id, rfc_slug, requester_user_id, matched_term, who_i_am, why,
+               use_case, status
+          FROM contribution_requests WHERE id = ? AND rfc_slug = ?
+        """,
+        (request_id, slug),
+    ).fetchone()
+    if row is None:
+        raise HTTPException(404, "Contribution request not found")
+    return row
+
+
+def _viewer_relationship(viewer, slug: str) -> str | None:
+    """Why this viewer can't *request* to contribute — or None if they can.
+    Owners/admins already have the RFC; existing collaborators are already
+    in. Both get a clear 409 rather than a useless self-request."""
+    if auth.is_rfc_owner(viewer, slug) or viewer.role in ("owner", "admin"):
+        return "You already own or administer this RFC."
+    if auth.is_rfc_collaborator(viewer, slug):
+        return "You're already a collaborator on this RFC."
+    return None
+
+
+def make_router() -> APIRouter:
+    router = APIRouter()
+
+    # ---------------------------------------------------------------
+    # GET — what the contribute form needs to render + gate itself.
+    # ---------------------------------------------------------------
+
+    @router.get("/api/rfcs/{slug}/contribution-target")
+    async def contribution_target(slug: str, request: Request) -> dict[str, Any]:
+        row = db.conn().execute(
+            "SELECT slug, title, state, owners_json, proposed_by FROM cached_rfcs WHERE slug = ?",
+            (slug,),
+        ).fetchone()
+        if row is None:
+            raise HTTPException(404, "RFC not found")
+
+        from . import rfc_links  # local import: avoid a module import cycle
+
+        owner = rfc_links._owner_display(db.conn(), row["owners_json"], row["proposed_by"])
+        viewer = auth.current_user(request)
+
+        eligible = True
+        reason: str | None = None
+        already_requested = False
+        if row["state"] != "super-draft":
+            eligible, reason = False, "This RFC is no longer pending."
+        elif viewer is None:
+            eligible, reason = False, "Sign in to ask to contribute."
+        elif viewer.permission_state != "granted":
+            eligible, reason = False, "Your beta access request is in review."
+        else:
+            reason = _viewer_relationship(viewer, slug)
+            if reason is not None:
+                eligible = False
+            else:
+                already_requested = bool(
+                    db.conn().execute(
+                        """
+                        SELECT 1 FROM contribution_requests
+                         WHERE rfc_slug = ? AND requester_user_id = ? AND status = 'pending'
+                         LIMIT 1
+                        """,
+                        (slug, viewer.user_id),
+                    ).fetchone()
+                )
+
+        return {
+            "slug": row["slug"],
+            "title": row["title"],
+            "owner": owner,
+            "eligible": eligible and not already_requested,
+            "reason": reason,
+            "already_requested": already_requested,
+        }
+
+    # ---------------------------------------------------------------
+    # POST — submit a contribute request.
+    # ---------------------------------------------------------------
+
+    @router.post("/api/rfcs/{slug}/contribution-requests")
+    async def create_contribution_request(
+        slug: str, body: ContributionRequestBody, request: Request
+    ) -> dict[str, Any]:
+        viewer = auth.require_contributor(request)
+        _require_super_draft(slug)
+
+        reason = _viewer_relationship(viewer, slug)
+        if reason is not None:
+            raise HTTPException(409, reason)
+
+        who_i_am = body.who_i_am.strip()
+        why = body.why.strip()
+        use_case = (body.use_case or "").strip() or None
+        matched_term = body.matched_term.strip()
+        if not who_i_am or not why:
+            raise HTTPException(422, "Both 'who I am' and 'why' are required.")
+
+        try:
+            cur = db.conn().execute(
+                """
+                INSERT INTO contribution_requests
+                    (rfc_slug, requester_user_id, matched_term, who_i_am, why, use_case)
+                VALUES (?, ?, ?, ?, ?, ?)
+                """,
+                (slug, viewer.user_id, matched_term, who_i_am, why, use_case),
+            )
+        except sqlite3.IntegrityError:
+            # The partial unique index — one open request per (RFC, user).
+            raise HTTPException(409, "You already have a pending request to contribute to this RFC.")
+        request_id = cur.lastrowid
+
+        # One actionable notification per owner; stamp the first onto the
+        # row as the inbox-action handle (any owner can act on the request).
+        notif_ids = notify.fan_out_contribution_request(
+            rfc_slug=slug,
+            requester_user_id=viewer.user_id,
+            request_id=request_id,
+            matched_term=matched_term,
+            who_i_am=who_i_am,
+            why=why,
+            use_case=use_case,
+        )
+        if notif_ids:
+            db.conn().execute(
+                "UPDATE contribution_requests SET notification_id = ? WHERE id = ?",
+                (notif_ids[0], request_id),
+            )
+
+        return {"id": request_id, "rfc_slug": slug, "status": "pending"}
+
+    # ---------------------------------------------------------------
+    # POST — owner accepts → fire #12's invite flow.
+    # ---------------------------------------------------------------
+
+    @router.post("/api/rfcs/{slug}/contribution-requests/{request_id}/accept")
+    async def accept_contribution_request(
+        slug: str, request_id: int, request: Request
+    ) -> dict[str, Any]:
+        viewer = auth.require_contributor(request)
+        rfc = _require_super_draft(slug)
+        if not auth.can_invite_to_rfc(viewer, slug):
+            raise HTTPException(403, "Only the RFC's owner can act on contribution requests")
+
+        req = _require_request(slug, request_id)
+        if req["status"] != "pending":
+            raise HTTPException(409, f"This request was already {req['status']}.")
+
+        requester = db.conn().execute(
+            "SELECT id, email FROM users WHERE id = ?", (req["requester_user_id"],)
+        ).fetchone()
+        if requester is None or not (requester["email"] or "").strip():
+            raise HTTPException(422, "The requester has no email address on file to invite.")
+
+        # Fire #12's owner-invite flow with the requester as the invitee.
+        # If a pending contributor invitation already exists (the owner
+        # invited them out-of-band first), reuse it rather than failing.
+        try:
+            invitation = api_invitations.issue_invitation(
+                slug=slug,
+                inviter_user_id=viewer.user_id,
+                inviter_display=viewer.display_name or viewer.gitea_login or "An RFC owner",
+                invitee_email=requester["email"],
+                role_in_rfc="contributor",
+                rfc_title=rfc["title"],
+            )
+            invitation_id = invitation["id"]
+        except HTTPException as exc:
+            if exc.status_code != 409:
+                raise
+            existing = db.conn().execute(
+                """
+                SELECT id FROM rfc_invitations
+                 WHERE rfc_slug = ? AND invitee_email = ? COLLATE NOCASE
+                   AND role_in_rfc = 'contributor' AND status = 'pending'
+                 ORDER BY id DESC LIMIT 1
+                """,
+                (slug, requester["email"].strip()),
+            ).fetchone()
+            invitation_id = existing["id"] if existing else None
+
+        db.conn().execute(
+            """
+            UPDATE contribution_requests
+               SET status = 'accepted', decided_at = datetime('now'),
+                   decided_by_user_id = ?, invitation_id = ?
+             WHERE id = ?
+            """,
+            (viewer.user_id, invitation_id, request_id),
+        )
+        notify.notify_contribution_decided(
+            rfc_slug=slug,
+            requester_user_id=req["requester_user_id"],
+            decider_user_id=viewer.user_id,
+            request_id=request_id,
+            accepted=True,
+        )
+        return {"ok": True, "status": "accepted", "invitation_id": invitation_id}
+
+    # ---------------------------------------------------------------
+    # POST — owner declines.
+    # ---------------------------------------------------------------
+
+    @router.post("/api/rfcs/{slug}/contribution-requests/{request_id}/decline")
+    async def decline_contribution_request(
+        slug: str, request_id: int, request: Request
+    ) -> dict[str, Any]:
+        viewer = auth.require_contributor(request)
+        _require_super_draft(slug)
+        if not auth.can_invite_to_rfc(viewer, slug):
+            raise HTTPException(403, "Only the RFC's owner can act on contribution requests")
+
+        req = _require_request(slug, request_id)
+        if req["status"] != "pending":
+            raise HTTPException(409, f"This request was already {req['status']}.")
+
+        db.conn().execute(
+            """
+            UPDATE contribution_requests
+               SET status = 'declined', decided_at = datetime('now'),
+                   decided_by_user_id = ?
+             WHERE id = ?
+            """,
+            (viewer.user_id, request_id),
+        )
+        notify.notify_contribution_decided(
+            rfc_slug=slug,
+            requester_user_id=req["requester_user_id"],
+            decider_user_id=viewer.user_id,
+            request_id=request_id,
+            accepted=False,
+        )
+        return {"ok": True, "status": "declined"}
+
+    return router
@@ -133,69 +133,15 @@ def make_router() -> APIRouter:
                "Only the RFC's owner can invite collaborators",
            )

-        invitee_email = body.invitee_email.strip()
-        role_in_rfc = body.role_in_rfc
-
-        # Refuse re-inviting an email that already has a pending
-        # invitation on this RFC at the same role. Different-role
-        # re-invite is allowed (upgrade discussant → contributor)
-        # — the new row supersedes the old in the UI listing's
-        # natural ordering, and acceptance of either picks up the
-        # corresponding role.
-        existing = db.conn().execute(
-            """
-            SELECT id FROM rfc_invitations
-             WHERE rfc_slug = ? AND invitee_email = ? COLLATE NOCASE
-               AND role_in_rfc = ? AND status = 'pending'
-             LIMIT 1
-            """,
-            (slug, invitee_email, role_in_rfc),
-        ).fetchone()
-        if existing:
-            raise HTTPException(
-                409,
-                f"{invitee_email} already has a pending {role_in_rfc} invitation for this RFC",
-            )
-
-        token = _mint_token()
-        cur = db.conn().execute(
-            """
-            INSERT INTO rfc_invitations
-                (rfc_slug, inviter_user_id, invitee_email, role_in_rfc,
-                 token, expires_at)
-            VALUES (?, ?, ?, ?, ?, datetime('now', ?))
-            """,
-            (
-                slug,
-                viewer.user_id,
-                invitee_email,
-                role_in_rfc,
-                token,
-                f"+{INVITATION_TTL_DAYS} days",
-            ),
-        )
-        invitation_id = cur.lastrowid
-
-        # Send the email — synchronous. A send failure logs and
-        # returns; the row stays so the owner can recover via the
-        # listing (which carries the token for an out-of-band share).
-        _send_invitation_email(
-            to_address=invitee_email,
+        return issue_invitation(
+            slug=slug,
+            inviter_user_id=viewer.user_id,
            inviter_display=viewer.display_name or viewer.gitea_login or "An RFC owner",
+            invitee_email=body.invitee_email,
+            role_in_rfc=body.role_in_rfc,
            rfc_title=rfc["title"],
-            role_in_rfc=role_in_rfc,
-            token=token,
        )

-        return {
-            "id": invitation_id,
-            "rfc_slug": slug,
-            "invitee_email": invitee_email,
-            "role_in_rfc": role_in_rfc,
-            "status": "pending",
-            "token": token,
-        }
-
    # ---------------------------------------------------------------
    # GET /api/rfcs/<slug>/invitations
    # The owner's listing of every invitation on the RFC, regardless
@@ -473,6 +419,78 @@ def _effective_status(row) -> str:
    return "expired" if is_past else "pending"


+def issue_invitation(
+    *,
+    slug: str,
+    inviter_user_id: int,
+    inviter_display: str,
+    invitee_email: str,
+    role_in_rfc: str,
+    rfc_title: str,
+) -> dict:
+    """Mint + persist + email one ``rfc_invitations`` row.
+
+    The single chokepoint for issuing an invitation: the owner's manual
+    `POST /api/rfcs/{slug}/invitations` endpoint and roadmap #28 Part 3's
+    accept path both route through here, so the dup-guard, token mint,
+    insert, and transactional email stay identical.
+
+    Refuses (409) re-inviting an email that already has a pending
+    invitation on this RFC at the same role. A different-role re-invite is
+    allowed (the discussant → contributor upgrade) — the new row
+    supersedes the old in the listing's natural ordering, and acceptance
+    of either picks up the corresponding role.
+
+    Returns the new row's dict (including the raw token, for the owner's
+    out-of-band share / the caller's record-keeping). A send failure logs
+    and returns; the row stays so the owner can recover via the listing.
+    """
+    invitee_email = invitee_email.strip()
+    existing = db.conn().execute(
+        """
+        SELECT id FROM rfc_invitations
+         WHERE rfc_slug = ? AND invitee_email = ? COLLATE NOCASE
+           AND role_in_rfc = ? AND status = 'pending'
+         LIMIT 1
+        """,
+        (slug, invitee_email, role_in_rfc),
+    ).fetchone()
+    if existing:
+        raise HTTPException(
+            409,
+            f"{invitee_email} already has a pending {role_in_rfc} invitation for this RFC",
+        )
+
+    token = _mint_token()
+    cur = db.conn().execute(
+        """
+        INSERT INTO rfc_invitations
+            (rfc_slug, inviter_user_id, invitee_email, role_in_rfc,
+             token, expires_at)
+        VALUES (?, ?, ?, ?, ?, datetime('now', ?))
+        """,
+        (slug, inviter_user_id, invitee_email, role_in_rfc, token, f"+{INVITATION_TTL_DAYS} days"),
+    )
+    invitation_id = cur.lastrowid
+
+    _send_invitation_email(
+        to_address=invitee_email,
+        inviter_display=inviter_display,
+        rfc_title=rfc_title,
+        role_in_rfc=role_in_rfc,
+        token=token,
+    )
+
+    return {
+        "id": invitation_id,
+        "rfc_slug": slug,
+        "invitee_email": invitee_email,
+        "role_in_rfc": role_in_rfc,
+        "status": "pending",
+        "token": token,
+    }
+
+
 def _mint_token() -> str:
    """A 256-bit URL-safe token. The token shape is opaque to the
    consumer; the email link encodes it as a query param."""
@@ -557,7 +557,26 @@ def make_router(config: Config) -> APIRouter:
        # stays unauthenticated for dev (the v1 contract).
        import os as _os
        expected = _os.environ.get("WEBHOOK_EMAIL_BOUNCE_SECRET", "").strip()
-        if expected:
+        # v0.25.0 (audit 0026 M5): fail closed. An unset secret used to
+        # leave this endpoint fully unauthenticated — anyone could suppress
+        # any user's mail by POSTing their address (email_opt_out_all flip
+        # below). Now an unset secret DISABLES the endpoint (503) instead
+        # of opening it. A dev that genuinely wants it open opts in
+        # explicitly with RFC_APP_INSECURE_BOUNCE_WEBHOOK=1, mirroring the
+        # RFC_APP_INSECURE_WEBHOOKS dev-bypass on the Gitea hook.
+        if not expected:
+            if _os.environ.get("RFC_APP_INSECURE_BOUNCE_WEBHOOK", "").strip() == "1":
+                log.warning(
+                    "email-bounce webhook running UNAUTHENTICATED "
+                    "(RFC_APP_INSECURE_BOUNCE_WEBHOOK=1) — never set this in production"
+                )
+            else:
+                log.error(
+                    "email-bounce webhook refused: WEBHOOK_EMAIL_BOUNCE_SECRET is unset "
+                    "(set the secret to enable, or RFC_APP_INSECURE_BOUNCE_WEBHOOK=1 for dev)"
+                )
+                raise HTTPException(503, "Bounce webhook not configured")
+        else:
            received = request.headers.get("X-Webhook-Secret", "")
            import hmac as _hmac
            if not received or not _hmac.compare_digest(expected, received):
@@ -219,6 +219,19 @@ async def refresh_rfc_repo(config: Config, gitea: Gitea, slug: str) -> None:
        open_pulls, closed_pulls = [], []
    for pull in open_pulls + closed_pulls:
        head_branch = pull.get("head", {}).get("ref", "")
+        # Same deleted-branch recovery as refresh_meta_pulls: a merged-and-
+        # deleted PR's `head.ref` collapses to `refs/pull/<N>/head`. Here
+        # the slug is known (param), so state still updates correctly and
+        # no ghost forms — but blindly storing the sentinel would clobber
+        # the real branch name api_prs.py relies on as a fallback ref when
+        # the merge commit is gone. Recover it from the stored row.
+        if not head_branch or head_branch.startswith("refs/pull/"):
+            prior = db.conn().execute(
+                "SELECT head_branch FROM cached_prs WHERE repo = ? AND pr_number = ?",
+                (repo_full, pull["number"]),
+            ).fetchone()
+            if prior and prior["head_branch"]:
+                head_branch = prior["head_branch"]
        state = _state_from_pull(pull)
        gitea_opener = (pull.get("user") or {}).get("login") or ""
        opened_by = _resolve_actor(
@@ -431,6 +444,25 @@ async def refresh_meta_pulls(config: Config, gitea: Gitea) -> None:

    for pull in open_pulls + closed_pulls:
        head_branch = pull.get("head", {}).get("ref", "")
+        # A merged-and-deleted PR's branch is no longer reported by Gitea
+        # as its real name — the `head.ref` collapses to the synthetic
+        # `refs/pull/<N>/head` sentinel (or empty). The slug + kind both
+        # derive from the branch name, so a deleted branch would parse to
+        # slug=None and the row would be skipped forever, freezing the
+        # cached_prs row at its last-seen `state='open'` — a permanent
+        # ghost "pending idea" for an entry that has actually merged
+        # (caught when the operator authoring lane in ROADMAP #35 merged
+        # an idea PR with the branch deleted; the web UX leaves branches
+        # in place so it never tripped this). Recover the original branch
+        # from the row we already stored when the PR was open — that row
+        # retains the real `head_branch` (migration 002).
+        if not head_branch or head_branch.startswith("refs/pull/"):
+            prior = db.conn().execute(
+                "SELECT head_branch FROM cached_prs WHERE repo = ? AND pr_number = ?",
+                (repo_full, pull["number"]),
+            ).fetchone()
+            if prior and prior["head_branch"]:
+                head_branch = prior["head_branch"]
        slug = _slug_from_head_branch(head_branch)
        if slug is None:
            continue
@@ -97,6 +97,13 @@ class IssueOutcome:
    raw_token: str
    row_id: int

+    @property
+    def cookie_value(self) -> str:
+        """The value to put in the `rfc_device_trust` cookie: the row-id
+        selector joined to the raw token (v0.25.0 / audit 0026 M1). The
+        selector lets `lookup` read one indexed row instead of scanning."""
+        return f"{self.row_id}.{self.raw_token}"
+

 def _new_token() -> str:
    return secrets.token_urlsafe(TOKEN_BYTES)
@@ -173,33 +180,37 @@ def lookup(raw_token: str) -> LookupOutcome:
    if not raw:
        return LookupOutcome(ok=False, user=None, reason="invalid")

-    # The unique index on `device_token_hash` would let us SELECT by
-    # hash if bcrypt were a stable hash, but bcrypt incorporates a
-    # per-row salt — equal tokens produce different hashes. We walk
-    # the candidate set instead. In practice the set is small (a
-    # human has a handful of trusted devices) and bcrypt is cheap on
-    # the order of milliseconds; the walk is bounded by the user's
-    # active device count.
+    # v0.25.0 (audit 0026 M1): the cookie is "<row_id>.<raw_token>". We
+    # parse the row-id selector and read exactly ONE row by its indexed
+    # primary key, then bcrypt-check the token against that single row.
    #
-    # We don't pre-filter by `revoked_at IS NULL` here so that a
-    # token presented for a recently-revoked row produces a
-    # 'revoked' outcome (the endpoint surfaces a different shape).
-    # Same for expired: we let the walk hit and classify after.
-    rows = db.conn().execute(
+    # The previous shape read EVERY device_trust row (all users, including
+    # revoked/expired) and bcrypt-checked each — an unauthenticated
+    # CPU-amplification DoS reachable at /auth/device-trust/start that
+    # grew without bound as the table accumulated. bcrypt's per-row salt
+    # is why we can't SELECT by hash; carrying the row-id in the cookie is
+    # the standard fix (the id is not secret; the token still is).
+    selector, sep, token = raw.partition(".")
+    if not sep or not selector.isdigit() or not token:
+        # Legacy bare-token cookies (pre-v0.25.0) and malformed values land
+        # here. We refuse rather than fall back to a full-table scan, so
+        # the amplification path is fully closed; affected users simply
+        # re-authenticate once via OTC/passcode and get a new cookie.
+        return LookupOutcome(ok=False, user=None, reason="invalid")
+
+    matched = db.conn().execute(
        """
        SELECT id, user_id, device_token_hash, expires_at, revoked_at
          FROM device_trust
-         ORDER BY id DESC
+         WHERE id = ?
        """,
-    ).fetchall()
+        (int(selector),),
+    ).fetchone()

-    matched = None
-    for row in rows:
-        if _check(raw, row["device_token_hash"]):
-            matched = row
-            break
-
-    if matched is None:
+    # One bcrypt check, against the selected row only. A wrong/forged token
+    # for a real id reads as 'unknown' (cookie cleared), same as a missing
+    # row — a probing client can't distinguish the two.
+    if matched is None or not _check(token, matched["device_token_hash"]):
        return LookupOutcome(ok=False, user=None, reason="unknown")

    if matched["revoked_at"] is not None:
@@ -60,12 +60,17 @@ def build_envelope(
    `from_name` is the display label that goes through `formataddr`
    so spaces / commas in the display string are encoded correctly.

-    `body_plain` is mandatory. `body_html`, if supplied, lands as the
-    second part of a `multipart/alternative` body — mail clients
-    that prefer HTML render it; clients that don't fall back to the
-    plain part. The text/plain part comes first per RFC 2046, so a
-    plain-text client that picks the first body gets the readable
-    text.
+    `body_plain` is mandatory. `body_html` is **reserved and not yet
+    enabled** (security-audit-0026 I3): no send path supplies it today —
+    every rfc-app mail is plain text — and passing it raises
+    `NotImplementedError`. The parameter is kept in the signature for
+    documented future symmetry: when HTML mail is enabled it will land
+    as the second part of a `multipart/alternative` body (text/plain
+    first per RFC 2046, so a plain-text client picking the first part
+    still gets the readable text). Enabling it is a deliberate act — the
+    caller MUST HTML-escape any user content into `body_html` first (cf.
+    the C1 stored-XSS class: a mail client renders the HTML) and remove
+    the guard below in the same change.

    `reply_to`, when set, lets a send path point replies at a
    different mailbox than the From line (e.g., a watcher
@@ -131,13 +136,20 @@ def build_envelope(
            # idempotent and not require auth. See
            # `api_notifications.py` for the receiver.
            msg["List-Unsubscribe-Post"] = "List-Unsubscribe=One-Click"
-    if body_html:
-        # multipart/alternative: text/plain first, text/html second.
-        # `set_content` sets the first part (and the message's main
-        # body); `add_alternative` adds the second part and
-        # restructures the message as multipart/alternative.
-        msg.set_content(body_plain)
-        msg.add_alternative(body_html, subtype="html")
-    else:
-        msg.set_content(body_plain)
+    if body_html is not None:
+        # I3 (security-audit-0026): the multipart/alternative HTML path
+        # is intentionally NOT enabled. No send path passes `body_html`
+        # today, and emitting an HTML body built from user-supplied
+        # content without escaping it first would reintroduce the C1
+        # stored-XSS class in the mail channel (the recipient's client
+        # renders the HTML). Fail loudly here rather than silently
+        # shipping HTML: enabling HTML mail is a deliberate change that
+        # MUST HTML-escape user content at the call site and remove this
+        # guard together. The text/plain path below is the only live one.
+        raise NotImplementedError(
+            "HTML email is not enabled (security-audit-0026 I3): do not "
+            "pass body_html until user content is HTML-escaped at the "
+            "call site and this guard is intentionally removed."
+        )
+    msg.set_content(body_plain)
    return msg
@@ -7,6 +7,7 @@ no need for a separate worker.
 from __future__ import annotations

 import logging
+import os
 import secrets
 from contextlib import asynccontextmanager

@@ -28,6 +29,7 @@ from . import (
    otc,
    passcode as passcode_mod,
    providers as providers_mod,
+    ratelimit,
    turnstile,
    webhooks,
 )
@@ -142,12 +144,20 @@ def create_app() -> FastAPI:
    # eagerly via load_config(). Everything else waits for lifespan.
    config = load_config()
    app = FastAPI(lifespan=lifespan)
+    # v0.25.0 (audit 0026 M4): the session cookie is the primary 30-day
+    # auth credential and must carry `Secure` in production so it never
+    # travels cleartext. Default to Secure; a dev box serving over plain
+    # http opts out with SESSION_COOKIE_SECURE=false. Production (OHM is
+    # HTTPS-only with an HTTP->HTTPS 301) leaves this unset → Secure on.
+    session_secure = os.environ.get("SESSION_COOKIE_SECURE", "true").strip().lower() not in (
+        "0", "false", "no", "off",
+    )
    app.add_middleware(
        SessionMiddleware,
        secret_key=config.secret_key,
        session_cookie="rfc_session",
        max_age=60 * 60 * 24 * 30,
-        https_only=False,
+        https_only=session_secure,
    )
    return app

@@ -155,24 +165,25 @@ def create_app() -> FastAPI:
 app = create_app()


-def _set_device_trust_cookie(response: Response, raw_token: str) -> None:
+def _set_device_trust_cookie(response: Response, cookie_value: str) -> None:
    """Attach the v0.11.0 device-trust cookie to the response.

-    HttpOnly + Secure + SameSite=Lax + 30-day Max-Age + Path=/. The
-    cookie value is the raw token; server-side storage is the hash.
-    The cookie is "essential" per the v0.13.0 cookie-consent contract
-    (it is part of authentication), so we set it regardless of the
-    user's analytics / other-cookies choice.
+    HttpOnly + Secure + SameSite=Lax + 30-day Max-Age + Path=/. As of
+    v0.25.0 (audit 0026 M1) the value is `IssueOutcome.cookie_value` —
+    "<row_id>.<raw_token>" — so `device_trust.lookup` can read one indexed
+    row instead of scanning; server-side storage remains the bcrypt hash
+    of the token half only. The cookie is "essential" per the v0.13.0
+    cookie-consent contract (it is part of authentication), so we set it
+    regardless of the user's analytics / other-cookies choice.

-    Secure=True means the cookie is only ever sent over HTTPS. The
-    SessionMiddleware in `create_app` keeps `https_only=False` for
-    dev parity, but the device-trust cookie holds a 30-day credential
-    and must not travel cleartext — production deployments serve over
-    HTTPS, so Secure on the device-trust cookie is non-negotiable.
+    Secure=True means the cookie is only ever sent over HTTPS — the
+    device-trust cookie holds a 30-day credential and must never travel
+    cleartext. (The session cookie now also defaults to Secure; see M4 in
+    `create_app`.)
    """
    response.set_cookie(
        key=device_trust_mod.COOKIE_NAME,
-        value=raw_token,
+        value=cookie_value,
        max_age=device_trust_mod.COOKIE_MAX_AGE_SECONDS,
        path="/",
        secure=True,
@@ -245,6 +256,10 @@ def _oauth_router(config) -> APIRouter:

    @router.post("/auth/otc/request")
    async def otc_request(body: OtcRequestBody, request: Request):
+        # v0.25.0 (audit 0026 H1/L2): per-IP brake at the cheapest point,
+        # before the Turnstile network call or any bcrypt/SMTP work.
+        if not ratelimit.otc_request_limiter.allow(ratelimit.client_key(request)):
+            raise HTTPException(429, "Too many requests; please wait a few minutes")
        # v0.12.0 / roadmap item #10: gate the request on a successful
        # Turnstile siteverify before the bcrypt hash + SMTP send. The
        # check runs first so a failed challenge spends no rate budget
@@ -252,7 +267,7 @@ def _oauth_router(config) -> APIRouter:
        # secret AND TURNSTILE_REQUIRED=false (the default), the gate
        # opens — see `backend/app/turnstile.py` for the full matrix.
        client_ip = request.client.host if request.client else None
-        ts = turnstile.verify_token(body.turnstile_token, client_ip=client_ip)
+        ts = await turnstile.verify_token(body.turnstile_token, client_ip=client_ip)
        if not ts.ok:
            if ts.reason == "misconfigured":
                # TURNSTILE_REQUIRED=true but the secret is unset. This
@@ -278,9 +293,25 @@ def _oauth_router(config) -> APIRouter:

    @router.post("/auth/otc/verify")
    async def otc_verify(body: OtcVerifyBody, request: Request, response: Response):
+        # v0.25.0 (audit 0026 H1): per-IP brake against fan-out guessing,
+        # plus the per-email lockout enforced inside otc.verify_code.
+        ip = ratelimit.client_key(request)
+        if not ratelimit.verify_limiter.allow(ip):
+            raise HTTPException(429, "Too many attempts; please wait a few minutes")
        result = otc.verify_code(body.email, body.code)
+        if result.reason == "locked":
+            raise HTTPException(
+                423,
+                {
+                    "detail": "Too many failed attempts; wait a few minutes or request a new code",
+                    "locked_until": result.locked_until,
+                },
+            )
        if not result.ok or result.user is None:
            raise HTTPException(400, "Invalid or expired code")
+        # Legit sign-in: clear this IP's window so a user who fat-fingered
+        # a couple of codes isn't left throttled.
+        ratelimit.verify_limiter.reset(ip)
        auth.store_session(request, result.user)
        # v0.8.0: surface `needs_profile` so the Login.jsx surface can
        # decide whether to advance to the first/last/why capture step
@@ -313,7 +344,7 @@ def _oauth_router(config) -> APIRouter:
        if body.trust_device:
            ua = request.headers.get("user-agent", "")
            outcome = device_trust_mod.issue(result.user.user_id, ua)
-            _set_device_trust_cookie(response, outcome.raw_token)
+            _set_device_trust_cookie(response, outcome.cookie_value)
        return {
            "ok": True,
            "user": {
@@ -337,12 +368,17 @@ def _oauth_router(config) -> APIRouter:
    # ---------------------------------------------------------------

    @router.get("/auth/passcode/check")
-    async def passcode_check(email: str = ""):
+    async def passcode_check(request: Request, email: str = ""):
        """Does this email have a passcode set? Anonymous endpoint —
        the Login.jsx flow calls this after the user types their email
        to decide whether to render a passcode input or fall back to
        OTC. We surface only the boolean; lockout state, the hash, and
-        the set-at stamp are not leaked here."""
+        the set-at stamp are not leaked here.
+
+        v0.25.0 (audit 0026 L3): per-IP rate limit so the has-passcode
+        boolean can't be bulk-harvested to enumerate accounts."""
+        if not ratelimit.check_limiter.allow(ratelimit.client_key(request)):
+            raise HTTPException(429, "Too many requests; please wait a few minutes")
        status = passcode_mod.passcode_status(email)
        return {"has_passcode": status.has_passcode}

@@ -376,6 +412,11 @@ def _oauth_router(config) -> APIRouter:
        v0.11.0: the body's `trust_device` flag, if true, mints a
        fresh device-trust row and sets the long-lived cookie. Same
        opt-in contract as `/auth/otc/verify`."""
+        # v0.25.0 (audit 0026 H1): per-IP brake in front of the per-account
+        # passcode lockout, so fan-out across emails is throttled too.
+        ip = ratelimit.client_key(request)
+        if not ratelimit.verify_limiter.allow(ip):
+            raise HTTPException(429, "Too many attempts; please wait a few minutes")
        result = passcode_mod.verify_passcode(body.email, body.passcode)
        if result.reason == "locked":
            raise HTTPException(
@@ -387,11 +428,12 @@ def _oauth_router(config) -> APIRouter:
            )
        if not result.ok or result.user is None:
            raise HTTPException(400, "Invalid passcode")
+        ratelimit.verify_limiter.reset(ip)
        auth.store_session(request, result.user)
        if body.trust_device:
            ua = request.headers.get("user-agent", "")
            outcome = device_trust_mod.issue(result.user.user_id, ua)
-            _set_device_trust_cookie(response, outcome.raw_token)
+            _set_device_trust_cookie(response, outcome.cookie_value)
        return {
            "ok": True,
            "user": {
@@ -459,7 +501,7 @@ def _oauth_router(config) -> APIRouter:
        if body.trust_device:
            ua = request.headers.get("user-agent", "")
            outcome = device_trust_mod.issue(result.user.user_id, ua)
-            _set_device_trust_cookie(response, outcome.raw_token)
+            _set_device_trust_cookie(response, outcome.cookie_value)

        # Has the user already set a passcode? (Could only happen via
        # an admin pre-population path that doesn't exist yet, but
@@ -270,6 +270,86 @@ def fan_out_new_beta_request(
        )


+def fan_out_contribution_request(
+    *,
+    rfc_slug: str,
+    requester_user_id: int,
+    request_id: int,
+    matched_term: str,
+    who_i_am: str,
+    why: str,
+    use_case: str | None,
+) -> list[int]:
+    """Roadmap #28 Part 3: a reader asked to contribute to a pending
+    (super-draft) RFC. Land one actionable notification per owner and
+    return their ids (the caller stamps the first onto the request row as
+    the inbox-action handle).
+
+    Personal-direct: the owner is the named subject of the request, so the
+    §15.4 email gate consults `email_personal_direct` exactly as for the
+    other owner-facing personal events — no new preference column is
+    needed. The request's three free-text fields ride along in the payload
+    so the inbox row can show the full ask inline without a second fetch.
+    Actor is the requester per §15.9.
+    """
+    requester = db.conn().execute(
+        "SELECT display_name FROM users WHERE id = ?", (requester_user_id,)
+    ).fetchone()
+    display = (requester["display_name"] if requester else None) or "Someone"
+    details = {
+        "request_id": request_id,
+        "matched_term": matched_term,
+        "requester_user_id": requester_user_id,
+        "requester_display": display,
+        "who_i_am": who_i_am,
+        "why": why,
+        "use_case": use_case or "",
+    }
+    notif_ids: list[int] = []
+    for recipient_id in _entry_owner_user_ids(rfc_slug):
+        if recipient_id == requester_user_id:
+            continue
+        notif_ids.append(
+            _emit_one(
+                recipient_user_id=recipient_id,
+                event_kind="contribution_request_on_pending_rfc",
+                category=CATEGORY_PERSONAL,
+                actor_user_id=requester_user_id,
+                rfc_slug=rfc_slug,
+                branch_name=None,
+                pr_number=None,
+                details=details,
+            )
+        )
+    return notif_ids
+
+
+def notify_contribution_decided(
+    *,
+    rfc_slug: str,
+    requester_user_id: int,
+    decider_user_id: int,
+    request_id: int,
+    accepted: bool,
+) -> None:
+    """Roadmap #28 Part 3: tell the requester an owner accepted or declined
+    their contribute request. On accept the requester also receives the
+    #12 invitation email out-of-band; this inbox row is the in-app echo
+    that points them at it."""
+    _emit_one(
+        recipient_user_id=requester_user_id,
+        event_kind=(
+            "contribution_request_accepted" if accepted else "contribution_request_declined"
+        ),
+        category=CATEGORY_PERSONAL,
+        actor_user_id=decider_user_id,
+        rfc_slug=rfc_slug,
+        branch_name=None,
+        pr_number=None,
+        details={"request_id": request_id},
+    )
+
+
 def fan_out_chat_message(
    *,
    actor_user_id: int,
@@ -769,6 +849,16 @@ def render_summary(event_kind: str, actor_display: str | None, rfc_title: str |
        return f"{actor} began graduating {title}."
    if event_kind == "pr_conflict_with_main":
        return f"{actor} started a resolution branch on {title}."
+    if event_kind == "contribution_request_on_pending_rfc":
+        # Roadmap #28 Part 3: owner-facing, actionable. The term is the
+        # super-draft reference that surfaced the offer; the inbox row
+        # renders Accept/Decline beneath this line.
+        term = extras.get("matched_term") or title
+        return f"{actor} wants to contribute to your pending RFC for '{term}'."
+    if event_kind == "contribution_request_accepted":
+        return f"{actor} accepted your request to contribute to {title} — check your email to accept the invitation."
+    if event_kind == "contribution_request_declined":
+        return f"{actor} declined your request to contribute to {title}."
    if event_kind == "new_beta_request":
        # v0.9.0: framework-scoped, not RFC-scoped. The actor (the
        # requester) and the captured full name + email read as
@@ -884,6 +974,11 @@ def list_inbox(
            "read_at": row["read_at"],
            "category": extras.get("category"),
            "summary": render_summary(row["event_kind"], row["actor_display"], row["rfc_title"], extras),
+            # The row's payload, surfaced for kinds that render inline
+            # detail (e.g. #28 Part 3's contribute-request who/why/use-case
+            # + Accept/Decline). Safe to expose: a recipient only ever sees
+            # their own notifications.
+            "extras": extras,
        })

    if bundled:
@@ -85,6 +85,15 @@ def _cooldown_seconds() -> int:
        return 60


+# v0.25.0 / security audit 0026 (H1): per-email OTC verify lockout,
+# mirroring the passcode path (passcode.py). Five consecutive wrong codes
+# for an email lock its OTC verify for 15 minutes. The per-IP limiter in
+# ratelimit.py is the primary brute-force brake; this is the durable,
+# passcode-parity layer.
+LOCKOUT_AFTER_FAILED_ATTEMPTS = 5
+LOCKOUT_DURATION_MINUTES = 15
+
+
 # ---------------------------------------------------------------------------
 # Code generation + hashing
 # ---------------------------------------------------------------------------
@@ -206,6 +215,61 @@ class VerifyOutcome:
    ok: bool
    user: SessionUser | None
    reason: str
+    # v0.25.0 (H1): ISO-8601 stamp when reason == 'locked'.
+    locked_until: str | None = None
+
+
+def _verify_lockout_until(email: str) -> str | None:
+    """Return the active lockout stamp for `email`, or None if not locked.
+
+    Clears an elapsed lockout (and resets the counter) as a side effect so
+    the next failure starts a fresh budget — mirrors passcode.verify_passcode.
+    """
+    row = db.conn().execute(
+        "SELECT failed_attempts, locked_until FROM otc_verify_state WHERE email = ?",
+        (email,),
+    ).fetchone()
+    if row is None or not row["locked_until"]:
+        return None
+    still_locked = db.conn().execute(
+        "SELECT datetime(?) > datetime('now') AS locked", (row["locked_until"],),
+    ).fetchone()["locked"]
+    if still_locked:
+        return row["locked_until"]
+    db.conn().execute(
+        "UPDATE otc_verify_state SET failed_attempts = 0, locked_until = NULL WHERE email = ?",
+        (email,),
+    )
+    return None
+
+
+def _record_verify_failure(email: str) -> None:
+    """Increment the per-email failure counter; stamp a lockout once it
+    crosses the threshold. Mirrors the passcode lockout shape."""
+    db.conn().execute(
+        """
+        INSERT INTO otc_verify_state (email, failed_attempts)
+        VALUES (?, 1)
+        ON CONFLICT(email) DO UPDATE SET failed_attempts = failed_attempts + 1
+        """,
+        (email,),
+    )
+    count = db.conn().execute(
+        "SELECT failed_attempts FROM otc_verify_state WHERE email = ?", (email,),
+    ).fetchone()["failed_attempts"]
+    if count >= LOCKOUT_AFTER_FAILED_ATTEMPTS:
+        db.conn().execute(
+            f"""
+            UPDATE otc_verify_state
+               SET locked_until = datetime('now', '+{LOCKOUT_DURATION_MINUTES} minutes')
+             WHERE email = ?
+            """,
+            (email,),
+        )
+
+
+def _clear_verify_state(email: str) -> None:
+    db.conn().execute("DELETE FROM otc_verify_state WHERE email = ?", (email,))


 def verify_code(email: str, code: str) -> VerifyOutcome:
@@ -214,6 +278,13 @@ def verify_code(email: str, code: str) -> VerifyOutcome:
    if not email or not code:
        return VerifyOutcome(ok=False, user=None, reason="invalid")

+    # v0.25.0 (H1): refuse before spending any bcrypt if this email is in
+    # its OTC-verify lockout window. The passcode path is unaffected — a
+    # locked-out OTC user can still set/use a passcode, and vice versa.
+    locked_until = _verify_lockout_until(email)
+    if locked_until:
+        return VerifyOutcome(ok=False, user=None, reason="locked", locked_until=locked_until)
+
    rows = db.conn().execute(
        """
        SELECT id, code_hash, expires_at, consumed_at
@@ -237,6 +308,9 @@ def verify_code(email: str, code: str) -> VerifyOutcome:
            break

    if matched is None:
+        # A genuine wrong guess against this email — the brute-force
+        # signal. Count it toward the lockout threshold (H1).
+        _record_verify_failure(email)
        return VerifyOutcome(ok=False, user=None, reason="wrong")

    if matched["consumed_at"] is not None:
@@ -255,6 +329,8 @@ def verify_code(email: str, code: str) -> VerifyOutcome:
        "UPDATE otc_codes SET consumed_at = datetime('now') WHERE id = ?",
        (matched["id"],),
    )
+    # Success wipes the per-email failure counter (H1).
+    _clear_verify_state(email)
    user = provision_or_link_user(email)
    return VerifyOutcome(ok=True, user=user, reason="ok")

@@ -0,0 +1,92 @@
+"""In-process per-IP sliding-window rate limiter (security audit 0026, H1).
+
+The auth verify endpoints (`/auth/otc/verify`, `/auth/passcode/verify`)
+had no per-IP brake, so an attacker could fan out guesses against a
+target identity bounded only by bcrypt cost. This module is the brake.
+
+It is deliberately tiny: §4.2 says the app is a single process with a
+colocated SQLite file, so an in-memory dict of `key -> deque[timestamps]`
+is sufficient and needs no shared store. State resets on restart, which
+fails *open* for a brief window — acceptable because the per-email OTC
+lockout (`otc_verify_state`) and the passcode lockout both persist in the
+database and carry the durable guarantee; this limiter is the
+anti-fan-out layer on top.
+
+Chosen over a per-identity lockout *as the primary control* because a
+per-IP window throttles the attacker without letting them grief a victim
+by locking that victim's account (the known downside of identity
+lockouts). Both layers run together.
+"""
+from __future__ import annotations
+
+import threading
+import time
+from collections import defaultdict, deque
+
+
+class SlidingWindowLimiter:
+    """Allow at most `max_events` per `window_seconds` per key.
+
+    `allow(key)` records an event and returns True if the key is still
+    within budget, False if it has exceeded it. Timestamps use a
+    monotonic clock so the limiter is immune to wall-clock jumps.
+    """
+
+    def __init__(self, max_events: int, window_seconds: float) -> None:
+        self.max_events = max_events
+        self.window_seconds = window_seconds
+        self._events: dict[str, deque[float]] = defaultdict(deque)
+        self._lock = threading.Lock()
+
+    def allow(self, key: str) -> bool:
+        now = time.monotonic()
+        cutoff = now - self.window_seconds
+        with self._lock:
+            q = self._events[key]
+            while q and q[0] < cutoff:
+                q.popleft()
+            if len(q) >= self.max_events:
+                return False
+            q.append(now)
+            # Opportunistic cleanup so idle keys don't accumulate forever.
+            if not q:
+                self._events.pop(key, None)
+            return True
+
+    def reset(self, key: str) -> None:
+        """Drop a key's window — e.g. after a successful sign-in so a
+        legitimate user who fat-fingered a few times isn't throttled."""
+        with self._lock:
+            self._events.pop(key, None)
+
+
+# Module-level limiters shared across requests (one process, so module
+# state is the natural home). Tunables are intentionally generous enough
+# not to bother a human retyping a code, tight enough to kill fan-out:
+#   * verify: 10 attempts / 5 min / IP across the auth verify surfaces.
+#   * otc request: 5 sends / 5 min / IP (Turnstile is the primary gate;
+#     this is defense in depth against a solved-challenge replay loop).
+verify_limiter = SlidingWindowLimiter(max_events=10, window_seconds=300)
+otc_request_limiter = SlidingWindowLimiter(max_events=5, window_seconds=300)
+# /auth/passcode/check is an anonymous has-passcode oracle (audit 0026 L3).
+# It's a legitimate Login-flow affordance, so the budget is generous —
+# enough for a human typing emails, tight enough to stop bulk scraping.
+check_limiter = SlidingWindowLimiter(max_events=30, window_seconds=300)
+
+
+def _reset_all_for_tests() -> None:
+    """Clear every module-level limiter's window. Test support only — the
+    limiters are process-global singletons, so without a per-test reset
+    one test's requests bleed into the next and later tests trip the
+    budget (429). Not called in production."""
+    for lim in (verify_limiter, otc_request_limiter, check_limiter):
+        with lim._lock:
+            lim._events.clear()
+
+
+def client_key(request) -> str:
+    """Best-effort client identity for limiting. Behind nginx the app is
+    started with `--forwarded-allow-ips 127.0.0.1`, so `request.client.host`
+    reflects the real client IP via Uvicorn's ProxyHeaders handling."""
+    client = getattr(request, "client", None)
+    return client.host if client and client.host else "unknown"
@@ -1,43 +1,95 @@
-"""Roadmap #28 Part 1 — auto-link RFC references in submitted prose.
+"""Roadmap #28 — scan submitted prose for RFC-shaped references.

-Scans plain-text PR descriptions and comment bodies for references to
-existing **accepted** (state='active') RFCs and returns a structured list
+The scanner splits a plain-text PR description / comment body into a list
 of *segments* the frontend renders: plain-text runs interleaved with
-``{"type": "rfc", ...}`` link segments. The backend never emits HTML —
-the frontend maps link segments onto React anchors — so the surface is
-XSS-safe by construction and independent of any HTML-sanitization layer.
+typed link segments. The backend never emits HTML — the frontend maps
+each segment onto a React node — so the surface is XSS-safe by
+construction and independent of any HTML-sanitization layer.

-**Read-time enrichment, not submit-time persistence.** The roadmap row
-phrases the scan as happening "at submit/post time"; this module instead
-enriches on read. The intent the roadmap actually names — "not as live
-compose preview" — is honored (drafts are never scanned, only submitted
-content on the read paths). Read-time was chosen for three reasons:
+Three buckets, one scan (Parts 1–3):

-  1. Correctness — links track the *live* active-RFC set. A newly-accepted
-     RFC starts linking in older comments; a withdrawn RFC stops linking
-     everywhere. Submit-time freezing would drift stale.
-  2. Zero migration — no derived data to store. (A concurrent session
-     already holds migration 023; staying migration-free keeps this slice
-     conflict-free as well as simpler.)
-  3. Cost — the active-RFC corpus is small and cache-resident, so building
-     the term index and scanning a ≤20k-char body per read is cheap.
+  * ``{"type": "rfc", ...}``          — Part 1. The term matches an
+    **accepted** (``state='active'``) RFC; renders as a link to it.
+  * ``{"type": "rfc-pending", ...}``  — Part 3. The term matches a
+    **pending** RFC — a super-draft (``state='super-draft'``: accepted
+    as an idea but not yet graduated to an active RFC) — which has an
+    owner and a contribution surface. Renders as an "ask to contribute"
+    affordance carrying the owner's display name.
+  * ``{"type": "rfc-candidate", ...}`` — Part 2. The term is a
+    strong-candidate that does **not** yet have a defining RFC. Renders
+    (for a viewer with create rights) as a "create RFC for '<term>'"
+    affordance that pre-fills the propose flow.

-**Matching is conservative by design.** Only references that are unlikely
-to be coincidental link:
+Precedence at any position is active > pending > candidate, then
+longest-match-first — an active link always wins over a contribute offer
+which always wins over a create offer for the same span.
+
+**Read-time enrichment, not submit-time persistence** (unchanged from
+Part 1): drafts are never scanned, only submitted content on the read
+paths, so links/offers track the *live* corpus. The active-RFC corpus,
+super-draft corpus, and tag taxonomy are all small and cache-resident,
+so building the index and scanning a ≤20k-char body per read is cheap.
+
+**Matching stays conservative by design.** A reference links/offers only
+when it is unlikely to be coincidental:

  * ``rfc_id`` tokens (e.g. ``RFC-0001``) — inherently specific.
  * Multi-word titles (containing whitespace, e.g. ``Open Human Model``).
  * Hyphenated slugs (containing ``-``, e.g. ``open-human-model``).

-Single common-word titles or slugs (e.g. a hypothetical RFC titled
-"Human") are deliberately NOT auto-linked — they would turn every prose
-"human" into a link. Surfacing those is the job of the roadmap's
-"curated canonical-terms list", an explicit per-deployment opt-in left as
-a future extension rather than guessed at here.
+Single common-word titles/slugs are deliberately NOT matched — they
+would turn every prose occurrence into an affordance.
+
+**Part 2 candidate heuristic.** A candidate term is a **multi-word tag**
+from the #27 tag taxonomy (the de-facto set of tags the corpus already
+carries) that has no defining RFC (no active or super-draft RFC whose
+slug or title is that term). Multi-word is the same false-positive guard
+the title rule uses: a single common tag word (``identity``) would be
+far too noisy. Broader candidate detection — capitalized multi-word
+phrases mined from the text, terms repeated across recently-touched PRs,
+or the #27 Haiku (``ANTHROPIC_API_KEY``) pathway — is a sanctioned but
+deferred extension; the conservative tag-taxonomy heuristic is chosen
+here to match Part 1's false-positive-averse philosophy.
 """
 from __future__ import annotations

-from typing import Any, Iterable
+import json
+import re
+from typing import Any, Iterable, NamedTuple
+
+
+class Term(NamedTuple):
+    """One match key plus what to emit when it hits.
+
+    ``key`` is the lowercase span to match (word-boundary, longest-first).
+    ``kind`` is ``'active' | 'pending' | 'candidate'`` and selects the
+    emitted segment shape. ``slug``/``title`` carry the target RFC (active
+    + pending); ``owner`` is the pending RFC's owner display name;
+    ``term`` is the candidate's canonical display spelling.
+    """
+
+    key: str
+    kind: str = "active"
+    slug: str = ""
+    title: str = ""
+    owner: str = ""
+    term: str = ""
+
+
+# Lower number = higher precedence when two keys of equal length match at
+# the same position. A real link beats a contribute offer beats a create
+# offer.
+_KIND_PRIORITY = {"active": 0, "pending": 1, "candidate": 2}
+
+
+def _coerce(t: Term | tuple) -> Term:
+    """Accept the legacy ``(key, slug, title)`` 3-tuple (treated as an
+    active term) alongside :class:`Term`, so direct unit-test callers and
+    older call sites keep working."""
+    if isinstance(t, Term):
+        return t
+    key, slug, title = t  # legacy active 3-tuple
+    return Term(key=key, kind="active", slug=slug, title=title)


 def _is_word_char(c: str) -> bool:
@@ -46,18 +98,37 @@ def _is_word_char(c: str) -> bool:
    return c.isalnum() or c in ("-", "_")


-def segment_text(text: str | None, terms: list[tuple[str, str, str]]) -> list[dict[str, Any]]:
-    """Split ``text`` into text / rfc-link segments against ``terms``.
+def _emit(term: Term, label: str) -> dict[str, Any]:
+    """The segment dict for a matched ``term``; ``label`` preserves source
+    casing."""
+    if term.kind == "pending":
+        return {
+            "type": "rfc-pending",
+            "slug": term.slug,
+            "label": label,
+            "title": term.title,
+            "owner": term.owner,
+        }
+    if term.kind == "candidate":
+        return {"type": "rfc-candidate", "label": label, "term": term.term}
+    return {"type": "rfc", "slug": term.slug, "label": label, "title": term.title}

-    ``terms`` is a list of ``(key_lower, slug, title)`` tuples; callers
-    pass it pre-sorted longest-first so the longest match wins at any
-    position (so "Open Human Model" wins over a bare "Open"). Matching is
-    case-insensitive and respects word boundaries on both ends. The
-    returned ``label`` preserves the source casing.
+
+def segment_text(text: str | None, terms: Iterable[Term | tuple]) -> list[dict[str, Any]]:
+    """Split ``text`` into text / link segments against ``terms``.
+
+    ``terms`` are :class:`Term` objects (or legacy ``(key, slug, title)``
+    active 3-tuples). Matching is case-insensitive, respects word
+    boundaries on both ends, and prefers the longest key — then higher
+    :data:`_KIND_PRIORITY` — at any position.

    Always returns at least one segment; for empty/None input that is a
    single empty text segment, so callers can render uniformly.
    """
+    ordered = sorted(
+        (_coerce(t) for t in terms),
+        key=lambda t: (-len(t.key), _KIND_PRIORITY.get(t.kind, 9)),
+    )
    if not text:
        return [{"type": "text", "text": text or ""}]

@@ -67,28 +138,23 @@ def segment_text(text: str | None, terms: list[tuple[str, str, str]]) -> list[di
    n = len(text)
    i = 0
    while i < n:
-        match: tuple[str, str, str, int] | None = None
-        for key, slug, title in terms:
-            klen = len(key)
-            if klen == 0 or not low.startswith(key, i):
+        match: tuple[Term, int] | None = None
+        for term in ordered:
+            klen = len(term.key)
+            if klen == 0 or not low.startswith(term.key, i):
                continue
            before = text[i - 1] if i > 0 else ""
            after = text[i + klen] if i + klen < n else ""
            if _is_word_char(before) or _is_word_char(after):
                continue
-            match = (key, slug, title, klen)
+            match = (term, klen)
            break
        if match is not None:
-            _key, slug, title, klen = match
+            term, klen = match
            if buf:
                out.append({"type": "text", "text": "".join(buf)})
                buf = []
-            out.append({
-                "type": "rfc",
-                "slug": slug,
-                "label": text[i:i + klen],
-                "title": title,
-            })
+            out.append(_emit(term, text[i:i + klen]))
            i += klen
        else:
            buf.append(text[i])
@@ -99,8 +165,8 @@ def segment_text(text: str | None, terms: list[tuple[str, str, str]]) -> list[di


 def _keys_for(slug: str, title: str, rfc_id: str | None) -> Iterable[str]:
-    """The match keys an active RFC contributes. See the module docstring
-    for why each gate exists (conservative, false-positive-averse)."""
+    """The match keys an RFC contributes. See the module docstring for why
+    each gate exists (conservative, false-positive-averse)."""
    if rfc_id:
        rid = rfc_id.strip()
        if len(rid) >= 2:
@@ -117,13 +183,23 @@ def _keys_for(slug: str, title: str, rfc_id: str | None) -> Iterable[str]:
            yield s.lower()


+def _slugify(term: str) -> str:
+    """Deterministic kebab-case — mirrors the propose modal's slugify so a
+    tag's would-be slug compares correctly against existing RFC slugs."""
+    return re.sub(r"-+$", "", re.sub(r"^-+", "", re.sub(r"[^a-z0-9]+", "-", term.lower().strip())))
+
+
 class LinkIndex:
    """A reusable term index built once per request and applied to many
    bodies (a PR's description plus every comment on it)."""

-    def __init__(self, terms: list[tuple[str, str, str]]):
-        # Longest key first so the longest reference wins at each position.
-        self._terms = sorted(terms, key=lambda t: len(t[0]), reverse=True)
+    def __init__(self, terms: Iterable[Term | tuple]):
+        # Coerce + order once; segment_text re-sorts defensively but a
+        # pre-sorted list keeps the per-body cost to the scan itself.
+        self._terms: list[Term] = sorted(
+            (_coerce(t) for t in terms),
+            key=lambda t: (-len(t.key), _KIND_PRIORITY.get(t.kind, 9)),
+        )

    def __bool__(self) -> bool:
        return bool(self._terms)
@@ -132,27 +208,126 @@ class LinkIndex:
        return segment_text(text, self._terms)


-def build_index(conn, *, exclude_slug: str | None = None) -> LinkIndex:
-    """Build a :class:`LinkIndex` from the accepted (active) RFC corpus.
+def _owner_display(conn, owners_json: str | None, proposed_by: str | None) -> str:
+    """The display name to show for a pending RFC's owner. First entry of
+    ``owners_json`` resolved to its user row's display name, falling back
+    to the bare login, then ``proposed_by``, then a neutral noun."""
+    login = None
+    try:
+        owners = json.loads(owners_json or "[]")
+        if isinstance(owners, list):
+            login = next((o for o in owners if isinstance(o, str) and o.strip()), None)
+    except (ValueError, TypeError):
+        login = None
+    if login:
+        row = conn.execute(
+            "SELECT display_name FROM users WHERE gitea_login = ?", (login,)
+        ).fetchone()
+        if row and row["display_name"]:
+            return row["display_name"]
+        return login
+    return (proposed_by or "").strip() or "the proposer"

-    ``exclude_slug`` drops the RFC the surrounding surface is itself scoped
-    to, so an RFC's own title/id/slug don't self-link inside its own PR or
-    discussion. ``ORDER BY slug`` makes key de-duplication deterministic
-    when two RFCs would contribute the same key (first slug wins)."""
-    rows = conn.execute(
-        "SELECT slug, title, rfc_id FROM cached_rfcs WHERE state = 'active' ORDER BY slug"
-    ).fetchall()
-    terms: list[tuple[str, str, str]] = []
+
+def _tag_universe(conn) -> list[str]:
+    """Distinct tags across the cached corpus (the #27 de-facto taxonomy),
+    preserving original spelling; case-deduped."""
+    rows = conn.execute("SELECT tags_json FROM cached_rfcs").fetchall()
+    out: list[str] = []
    seen: set[str] = set()
    for r in rows:
+        try:
+            tags = json.loads(r["tags_json"] or "[]")
+        except (ValueError, TypeError):
+            continue
+        if not isinstance(tags, list):
+            continue
+        for t in tags:
+            if not isinstance(t, str):
+                continue
+            tag = t.strip()
+            low = tag.lower()
+            if tag and low not in seen:
+                seen.add(low)
+                out.append(tag)
+    return out
+
+
+def build_index(
+    conn,
+    *,
+    exclude_slug: str | None = None,
+    include_pending: bool = True,
+    include_candidates: bool = True,
+) -> LinkIndex:
+    """Build a :class:`LinkIndex` over the three buckets.
+
+    ``exclude_slug`` drops the RFC the surrounding surface is itself scoped
+    to, so an RFC's own title/id/slug don't self-link (or self-offer)
+    inside its own PR or discussion. Precedence is enforced by insertion
+    order — active keys are added first and a later bucket never overrides
+    an already-claimed key.
+    """
+    terms: list[Term] = []
+    seen: set[str] = set()
+
+    def add(key: str, term: Term) -> None:
+        if key in seen:
+            return
+        seen.add(key)
+        terms.append(term)
+
+    # --- Part 1: accepted (active) RFCs. ORDER BY slug makes key
+    # de-duplication deterministic when two RFCs would contribute the
+    # same key (first slug wins). ---
+    active_rows = conn.execute(
+        "SELECT slug, title, rfc_id FROM cached_rfcs WHERE state = 'active' ORDER BY slug"
+    ).fetchall()
+    # Track every slug + title that *has* a defining RFC, so Part 2 never
+    # offers to create one that already exists (active or pending).
+    defined_slugs: set[str] = set()
+    defined_titles: set[str] = set()
+    for r in active_rows:
        slug = r["slug"]
+        defined_slugs.add((slug or "").lower())
+        defined_titles.add((r["title"] or "").strip().lower())
        if exclude_slug is not None and slug == exclude_slug:
            continue
        title = r["title"] or ""
        rfc_id = r["rfc_id"] if "rfc_id" in r.keys() else None
        for key in _keys_for(slug, title, rfc_id):
-            if key in seen:
+            add(key, Term(key=key, kind="active", slug=slug, title=title))
+
+    # --- Part 3: pending (super-draft) RFCs. ---
+    pending_rows = conn.execute(
+        """
+        SELECT slug, title, rfc_id, owners_json, proposed_by
+          FROM cached_rfcs WHERE state = 'super-draft' ORDER BY slug
+        """
+    ).fetchall()
+    for r in pending_rows:
+        slug = r["slug"]
+        defined_slugs.add((slug or "").lower())
+        defined_titles.add((r["title"] or "").strip().lower())
+        if not include_pending:
+            continue
+        if exclude_slug is not None and slug == exclude_slug:
+            continue
+        title = r["title"] or ""
+        rfc_id = r["rfc_id"] if "rfc_id" in r.keys() else None
+        owner = _owner_display(conn, r["owners_json"], r["proposed_by"])
+        for key in _keys_for(slug, title, rfc_id):
+            add(key, Term(key=key, kind="pending", slug=slug, title=title, owner=owner))
+
+    # --- Part 2: strong-candidate terms with no defining RFC. ---
+    if include_candidates:
+        for tag in _tag_universe(conn):
+            low = tag.lower()
+            # Conservative: multi-word tags only (same guard as titles).
+            if " " not in tag and "\t" not in tag:
                continue
-            seen.add(key)
-            terms.append((key, slug, title))
+            if low in defined_titles or low in defined_slugs or _slugify(tag) in defined_slugs:
+                continue
+            add(low, Term(key=low, kind="candidate", term=tag))
+
    return LinkIndex(terms)
@@ -73,6 +73,19 @@ def _siteverify_url() -> str:
    return os.environ.get("TURNSTILE_SITEVERIFY_URL", "").strip() or SITEVERIFY_URL


+async def _siteverify_post(url: str, data: dict) -> httpx.Response:
+    """Perform the siteverify POST on an `httpx.AsyncClient`.
+
+    Isolated as a narrow seam (I4, security-audit-0026): the call is
+    awaited so a slow CloudFlare response can't block the event loop,
+    and tests patch *this function* rather than the shared
+    `httpx.AsyncClient` (which other modules — gitea, docs — also
+    construct, so a global patch would break app boot).
+    """
+    async with httpx.AsyncClient(timeout=10.0) as client:
+        return await client.post(url, data=data)
+
+
@dataclass
 class VerifyOutcome:
    """Result of a Turnstile siteverify call.
@@ -98,7 +111,7 @@ class VerifyOutcome:
    reason: str


-def verify_token(token: str | None, *, client_ip: str | None = None) -> VerifyOutcome:
+async def verify_token(token: str | None, *, client_ip: str | None = None) -> VerifyOutcome:
    """Validate a Turnstile token against CloudFlare's siteverify endpoint.

    Returns a VerifyOutcome describing whether the calling endpoint
@@ -108,9 +121,14 @@ def verify_token(token: str | None, *, client_ip: str | None = None) -> VerifyOu
      * 'misconfigured' → 500 "auth misconfigured"
      * 'missing-token' / 'failed' / 'network' → 400 "verification failed"

-    Tests monkeypatch `httpx.post` (or set `TURNSTILE_SITEVERIFY_URL`
-    + a MockTransport client) to avoid touching the real CloudFlare
-    endpoint. No real keys are ever embedded in tests.
+    Async (I4, security-audit-0026): the siteverify call is awaited on an
+    `httpx.AsyncClient` so a slow CloudFlare response can't block the
+    event loop (the prior synchronous `httpx.post` stalled the single
+    worker for up to the 10s timeout). Callers must `await` it.
+
+    Tests monkeypatch `_siteverify_post` (the narrow async seam) to avoid
+    touching the real CloudFlare endpoint and to keep the patch off the
+    shared `httpx.AsyncClient`. No real keys are ever embedded in tests.
    """
    secret = _secret()
    required = _required()
@@ -133,7 +151,7 @@ def verify_token(token: str | None, *, client_ip: str | None = None) -> VerifyOu
        data["remoteip"] = client_ip

    try:
-        response = httpx.post(_siteverify_url(), data=data, timeout=10.0)
+        response = await _siteverify_post(_siteverify_url(), data)
        payload = response.json()
    except Exception as exc:  # network, JSON parse, etc.
        log.warning("Turnstile siteverify call failed: %s", exc)
@@ -0,0 +1,18 @@
+-- v0.25.0 / security audit 0026, finding H1.
+--
+-- The OTC verify path had no attempt-limit or lockout, unlike the
+-- passcode path (015_passcode.sql gave users.passcode_failed_attempts +
+-- passcode_locked_until). This table gives the OTC verify endpoint the
+-- same per-identity lockout shape. It is keyed by email rather than
+-- user_id because an OTC sign-in may not have a users row yet — the row
+-- is provisioned only on a *successful* verify, so the lockout state has
+-- to survive independently of it.
+--
+-- The per-IP rate limiter (app/ratelimit.py) is the primary brute-force
+-- defense; this table is the parity layer that mirrors the passcode
+-- lockout and persists across restarts.
+CREATE TABLE IF NOT EXISTS otc_verify_state (
+    email TEXT PRIMARY KEY,
+    failed_attempts INTEGER NOT NULL DEFAULT 0,
+    locked_until TEXT
+);
@@ -0,0 +1,59 @@
+-- v0.29.0 / roadmap #28 Part 3 — offer-to-contribute-to-a-pending-RFC.
+--
+-- When the #28 scanner matches a term in submitted PR/comment text to a
+-- *pending* RFC (a super-draft: accepted-as-an-idea but not yet graduated
+-- to an active RFC), the reader is offered a "ask to contribute" popover.
+-- Submitting it lands a row here AND a notification in each owner's §15
+-- inbox; the owner can accept (which fires #12's owner-invite flow with
+-- the requester as the invitee) or decline (the requester is notified and
+-- the request closes).
+--
+-- A "pending RFC" is scoped to a super-draft (cached_rfcs.state =
+-- 'super-draft'): it is in cached_rfcs (so the rfc_invitations FK that the
+-- accept path reuses resolves), it carries owners (owners_json) to route
+-- the request to, and it already has a discussion/contribution surface to
+-- open. Pre-merge idea PRs (not yet in cached_rfcs, no contribution
+-- surface) are deliberately out of scope — see backend/app/rfc_links.py.
+--
+-- The request row is the persistent record; the inbox notification is the
+-- owner-facing actionable surface keyed back to it via `notification_id`.
+
+CREATE TABLE IF NOT EXISTS contribution_requests (
+    id                  INTEGER PRIMARY KEY AUTOINCREMENT,
+    rfc_slug            TEXT NOT NULL
+                            REFERENCES cached_rfcs(slug) ON DELETE CASCADE,
+    requester_user_id   INTEGER NOT NULL
+                            REFERENCES users(id) ON DELETE CASCADE,
+    -- The term in the PR/comment text that surfaced the offer (e.g. the
+    -- super-draft's title). Carried for the owner's context line and the
+    -- requester's "what RFC" anchor; not a foreign key.
+    matched_term        TEXT NOT NULL,
+    -- The three contribute-request fields (§15 / #26 vocabulary).
+    -- `who_i_am` and `why` are required; `use_case` mirrors #26's
+    -- optional ground-truth field.
+    who_i_am            TEXT NOT NULL,
+    why                 TEXT NOT NULL,
+    use_case            TEXT,
+    status              TEXT NOT NULL DEFAULT 'pending'
+                            CHECK (status IN ('pending', 'accepted', 'declined')),
+    created_at          TEXT NOT NULL DEFAULT (datetime('now')),
+    decided_at          TEXT,
+    decided_by_user_id  INTEGER REFERENCES users(id) ON DELETE SET NULL,
+    -- The rfc_invitations row minted on accept (the #12 reuse), and the
+    -- owner-facing notification row that carries the Accept/Decline action.
+    invitation_id       INTEGER REFERENCES rfc_invitations(id) ON DELETE SET NULL,
+    notification_id     INTEGER REFERENCES notifications(id) ON DELETE SET NULL
+);
+
+CREATE INDEX IF NOT EXISTS idx_contribution_requests_rfc
+    ON contribution_requests(rfc_slug, status);
+
+CREATE INDEX IF NOT EXISTS idx_contribution_requests_requester
+    ON contribution_requests(requester_user_id, status);
+
+-- At most one open (pending) request per (RFC, requester): a second ask
+-- while one is still pending is a 409, not a duplicate row. A decided
+-- request (accepted/declined) does not block a fresh ask later.
+CREATE UNIQUE INDEX IF NOT EXISTS idx_contribution_requests_one_open
+    ON contribution_requests(rfc_slug, requester_user_id)
+    WHERE status = 'pending';
@@ -0,0 +1,19 @@
+"""Shared pytest fixtures for the backend suite.
+
+Added in v0.27.0 (security audit 0026) alongside the new per-IP rate
+limiter. The limiters in `app.ratelimit` are process-global singletons,
+so their state survives across tests within a run; without a reset, the
+accumulated requests from earlier tests exhaust the budget and later
+tests see spurious 429s. This autouse fixture gives every test a clean
+limiter window.
+"""
+import pytest
+
+from app import ratelimit
+
+
+@pytest.fixture(autouse=True)
+def _reset_rate_limiters():
+    ratelimit._reset_all_for_tests()
+    yield
+    ratelimit._reset_all_for_tests()
@@ -0,0 +1,236 @@
+"""v0.29.0 / roadmap #28 Parts 2 & 3 — create-RFC offers + contribute-to-
+pending requests.
+
+Two layers, mirroring test_rfc_links_vertical.py:
+
+  * The PR-view scanner surfaces `rfc-pending` (Part 3) and `rfc-candidate`
+    (Part 2) segments alongside Part 1's `rfc` links.
+  * The contribute-request flow end-to-end: a non-owner asks, each owner
+    gets an actionable §15 notification, accept fires #12's invite flow,
+    decline notifies the requester.
+
+Reuses the FakeGitea + seed/session helpers from the existing suites.
+"""
+from __future__ import annotations
+
+import json
+
+from fastapi.testclient import TestClient
+
+from app import db
+
+from test_propose_vertical import (  # noqa: F401
+    FakeGitea,
+    app_with_fake_gitea,
+    provision_user_row,
+    sign_in_as,
+    tmp_env,
+)
+from test_rfc_view_vertical import SEED_BODY, seed_active_rfc
+from test_super_draft_vertical import seed_super_draft
+from test_rfc_links_vertical import _open_pr_on
+
+
+def _set_owner(slug: str, login: str) -> None:
+    db.conn().execute(
+        "UPDATE cached_rfcs SET owners_json = ? WHERE slug = ?",
+        (json.dumps([login]), slug),
+    )
+
+
+def _set_tags(slug: str, tags: list[str]) -> None:
+    db.conn().execute(
+        "UPDATE cached_rfcs SET tags_json = ? WHERE slug = ?",
+        (json.dumps(tags), slug),
+    )
+
+
+def _segs(segments, kind):
+    return [s for s in segments if s["type"] == kind]
+
+
+# ---------------------------------------------------------------------------
+# Part 2 + Part 3 — scanner surfaces on the PR view
+# ---------------------------------------------------------------------------
+
+
+def test_pending_and_candidate_segments_on_pr(app_with_fake_gitea):
+    app, fake = app_with_fake_gitea
+    with TestClient(app) as client:
+        provision_user_row(user_id=2, login="alice", role="contributor")
+        # Host active RFC (OHM — single word, contributes no keys itself)
+        # carrying a multi-word tag with no defining RFC: the Part 2
+        # candidate. And a pending super-draft owned by alice: the Part 3
+        # contribute target.
+        seed_active_rfc(fake, slug="ohm", title="OHM", body=SEED_BODY)
+        _set_tags("ohm", ["memory model", "identity"])
+        seed_super_draft(fake, slug="open-human-model", title="Open Human Model",
+                         pitch="A framework for representing humans.", proposed_by="alice")
+        _set_owner("open-human-model", "alice")
+        sign_in_as(client, user_id=2, gitea_login="alice", display_name="Alice", role="contributor")
+
+        pr_number = _open_pr_on(
+            client, fake, host_slug="ohm",
+            description="This builds on the Open Human Model and the memory model.",
+        )
+        pr = client.get(f"/api/rfcs/ohm/prs/{pr_number}").json()
+        segs = pr["description_segments"]
+
+        pending = _segs(segs, "rfc-pending")
+        assert len(pending) == 1
+        assert pending[0]["slug"] == "open-human-model"
+        assert pending[0]["label"] == "Open Human Model"
+        assert pending[0]["owner"] == "Alice"   # display_name of the owner
+
+        candidate = _segs(segs, "rfc-candidate")
+        assert len(candidate) == 1
+        assert candidate[0]["term"] == "memory model"
+        # "identity" is a single-word tag — deliberately NOT a candidate.
+        assert all("identity" not in s.get("term", "") for s in candidate)
+
+
+# ---------------------------------------------------------------------------
+# Part 3 — the contribute-request flow
+# ---------------------------------------------------------------------------
+
+
+def _seed_pending_owned_by_alice(fake):
+    provision_user_row(user_id=2, login="alice", role="contributor")
+    provision_user_row(user_id=3, login="bob", role="contributor")
+    seed_super_draft(fake, slug="open-human-model", title="Open Human Model",
+                     pitch="A framework.", proposed_by="alice")
+    _set_owner("open-human-model", "alice")
+
+
+_REQUEST = {
+    "matched_term": "Open Human Model",
+    "who_i_am": "Bob, a researcher",
+    "why": "I have relevant prior work to bring.",
+    "use_case": "Building an identity tool.",
+}
+
+
+def test_request_accept_invites_and_notifies(app_with_fake_gitea):
+    app, fake = app_with_fake_gitea
+    with TestClient(app) as client:
+        _seed_pending_owned_by_alice(fake)
+
+        # Bob asks to contribute.
+        sign_in_as(client, user_id=3, gitea_login="bob", display_name="Bob",
+                   role="contributor", email="bob@test")
+        r = client.post("/api/rfcs/open-human-model/contribution-requests", json=_REQUEST)
+        assert r.status_code == 200, r.text
+        request_id = r.json()["id"]
+        assert r.json()["status"] == "pending"
+
+        # A second ask while pending is a 409, not a duplicate row.
+        assert client.post("/api/rfcs/open-human-model/contribution-requests",
+                           json=_REQUEST).status_code == 409
+
+        # Alice (owner) sees the actionable notification with full detail.
+        sign_in_as(client, user_id=2, gitea_login="alice", display_name="Alice",
+                   role="contributor", email="alice@test")
+        inbox = client.get("/api/notifications").json()
+        reqs = [i for i in inbox["items"]
+                if i["event_kind"] == "contribution_request_on_pending_rfc"]
+        assert len(reqs) == 1
+        assert "wants to contribute" in reqs[0]["summary"]
+        assert reqs[0]["extras"]["who_i_am"] == "Bob, a researcher"
+        assert reqs[0]["extras"]["request_id"] == request_id
+
+        # Alice accepts → #12 invitation minted for bob's email.
+        acc = client.post(f"/api/rfcs/open-human-model/contribution-requests/{request_id}/accept")
+        assert acc.status_code == 200, acc.text
+        assert acc.json()["status"] == "accepted"
+        assert acc.json()["invitation_id"]
+
+        inv = db.conn().execute(
+            "SELECT invitee_email, role_in_rfc, status FROM rfc_invitations "
+            "WHERE rfc_slug = 'open-human-model'"
+        ).fetchone()
+        assert inv["invitee_email"] == "bob@test"
+        assert inv["role_in_rfc"] == "contributor"
+        assert inv["status"] == "pending"
+
+        # The request is settled — re-accepting is a 409.
+        assert client.post(
+            f"/api/rfcs/open-human-model/contribution-requests/{request_id}/accept"
+        ).status_code == 409
+
+        # Bob gets the accepted echo in his inbox.
+        sign_in_as(client, user_id=3, gitea_login="bob", display_name="Bob",
+                   role="contributor", email="bob@test")
+        bob_kinds = [i["event_kind"] for i in client.get("/api/notifications").json()["items"]]
+        assert "contribution_request_accepted" in bob_kinds
+
+
+def test_decline_notifies_requester(app_with_fake_gitea):
+    app, fake = app_with_fake_gitea
+    with TestClient(app) as client:
+        _seed_pending_owned_by_alice(fake)
+        sign_in_as(client, user_id=3, gitea_login="bob", display_name="Bob",
+                   role="contributor", email="bob@test")
+        request_id = client.post(
+            "/api/rfcs/open-human-model/contribution-requests", json=_REQUEST
+        ).json()["id"]
+
+        sign_in_as(client, user_id=2, gitea_login="alice", display_name="Alice",
+                   role="contributor", email="alice@test")
+        dec = client.post(f"/api/rfcs/open-human-model/contribution-requests/{request_id}/decline")
+        assert dec.status_code == 200, dec.text
+        assert dec.json()["status"] == "declined"
+
+        row = db.conn().execute(
+            "SELECT status FROM contribution_requests WHERE id = ?", (request_id,)
+        ).fetchone()
+        assert row["status"] == "declined"
+
+        sign_in_as(client, user_id=3, gitea_login="bob", display_name="Bob",
+                   role="contributor", email="bob@test")
+        bob_kinds = [i["event_kind"] for i in client.get("/api/notifications").json()["items"]]
+        assert "contribution_request_declined" in bob_kinds
+
+
+def test_owner_cannot_request_own_rfc(app_with_fake_gitea):
+    app, fake = app_with_fake_gitea
+    with TestClient(app) as client:
+        _seed_pending_owned_by_alice(fake)
+        sign_in_as(client, user_id=2, gitea_login="alice", display_name="Alice",
+                   role="contributor", email="alice@test")
+        r = client.post("/api/rfcs/open-human-model/contribution-requests", json=_REQUEST)
+        assert r.status_code == 409
+        assert "own" in r.json()["detail"].lower()
+
+
+def test_request_on_active_rfc_rejected(app_with_fake_gitea):
+    # The contribute offer only exists for pending super-drafts; an active
+    # RFC uses the Part-1 link instead.
+    app, fake = app_with_fake_gitea
+    with TestClient(app) as client:
+        provision_user_row(user_id=3, login="bob", role="contributor")
+        seed_active_rfc(fake, slug="ohm", title="OHM", body=SEED_BODY)
+        sign_in_as(client, user_id=3, gitea_login="bob", display_name="Bob",
+                   role="contributor", email="bob@test")
+        r = client.post("/api/rfcs/ohm/contribution-requests", json=_REQUEST)
+        assert r.status_code == 409
+
+
+def test_contribution_target_eligibility(app_with_fake_gitea):
+    app, fake = app_with_fake_gitea
+    with TestClient(app) as client:
+        _seed_pending_owned_by_alice(fake)
+
+        # Anonymous: not eligible, told to sign in.
+        t = client.get("/api/rfcs/open-human-model/contribution-target").json()
+        assert t["eligible"] is False
+        assert "sign in" in (t["reason"] or "").lower()
+        assert t["owner"] == "Alice"
+
+        # Bob: eligible until he has a pending ask, then not.
+        sign_in_as(client, user_id=3, gitea_login="bob", display_name="Bob",
+                   role="contributor", email="bob@test")
+        assert client.get("/api/rfcs/open-human-model/contribution-target").json()["eligible"] is True
+        client.post("/api/rfcs/open-human-model/contribution-requests", json=_REQUEST)
+        after = client.get("/api/rfcs/open-human-model/contribution-target").json()
+        assert after["already_requested"] is True
+        assert after["eligible"] is False
@@ -11,6 +11,8 @@ from __future__ import annotations

 from email.utils import parsedate_to_datetime

+import pytest
+
 from app.email_envelope import build_envelope


@@ -160,14 +162,18 @@ def test_envelope_plain_only_body_is_text_plain():
    assert msg.get_content().strip() == "Hello, world."


-def test_envelope_with_html_is_multipart_alternative():
-    msg = build_envelope(**_base_kwargs(body_html="<p>Hello, <b>world</b>.</p>"))
-    assert msg.get_content_type() == "multipart/alternative"
-    # Two parts: text/plain first (so plain-text clients picking the
-    # first part get the readable text), text/html second.
-    parts = list(msg.iter_parts())
-    assert len(parts) == 2
-    assert parts[0].get_content_type() == "text/plain"
-    assert parts[1].get_content_type() == "text/html"
-    assert "Hello, world." in parts[0].get_content()
-    assert "<b>world</b>" in parts[1].get_content()
+def test_envelope_html_body_is_guarded_not_enabled():
+    # I3 (security-audit-0026): the HTML/multipart-alternative path is
+    # intentionally not enabled — passing body_html must fail loudly so
+    # a future caller can't silently ship unescaped user HTML (the C1
+    # stored-XSS class in the mail channel). When HTML mail is enabled
+    # deliberately, this test flips to assert the multipart shape.
+    with pytest.raises(NotImplementedError):
+        build_envelope(**_base_kwargs(body_html="<p>Hello, <b>world</b>.</p>"))
+
+
+def test_envelope_html_none_is_plain_only():
+    # The guard keys on `is not None`, so the default (None) stays the
+    # live plain-text path — exercised here to lock the boundary.
+    msg = build_envelope(**_base_kwargs(body_html=None))
+    assert msg.get_content_type() == "text/plain"
@@ -444,6 +444,18 @@ def tmp_env(monkeypatch):
        # the dev-bypass path monkeypatch `RFC_APP_INSECURE_WEBHOOKS=1`.
        "GITEA_WEBHOOK_SECRET": "test-webhook-secret-for-signature-verification",
        "ENABLED_MODELS": "claude",
+        # v0.27.0 (audit 0026 M4): the session cookie now defaults to
+        # Secure. The TestClient talks plain http://testserver, so a
+        # Secure cookie is never sent back and every authenticated flow
+        # would fail. Tests opt out explicitly, exactly as a dev box on
+        # plain http does.
+        "SESSION_COOKIE_SECURE": "false",
+        # v0.27.0 (audit 0026 M5): the bounce webhook fails closed (503)
+        # when its secret is unset. Tests exercise the legacy behavioral
+        # path via the documented dev opt-in, mirroring the
+        # RFC_APP_INSECURE_WEBHOOKS bypass above. Tests that assert the
+        # fail-closed default delenv this key themselves.
+        "RFC_APP_INSECURE_BOUNCE_WEBHOOK": "1",
    }
    for k, v in env.items():
        monkeypatch.setenv(k, v)
@@ -565,6 +577,97 @@ def test_propose_to_super_draft_vertical(app_with_fake_gitea):
        assert ("merge_proposal", "ben") in kinds


+def test_merged_idea_pr_with_deleted_branch_clears_proposal(app_with_fake_gitea):
+    """Regression: a merged idea PR whose branch was deleted must not
+    linger as a 'pending idea' ghost.
+
+    Found via the ROADMAP #35 operator authoring lane: merging an idea
+    PR from the CLI with `--delete-branch` makes Gitea report the PR's
+    `head.ref` as the synthetic `refs/pull/<N>/head` sentinel instead of
+    `propose/<slug>`. `refresh_meta_pulls` derives the slug from the
+    branch name, so the sentinel parsed to slug=None, the row was skipped,
+    and `cached_prs.state` stayed frozen at 'open' — leaving the entry
+    showing as BOTH a super-draft (cached_rfcs reconciled off the push)
+    AND a pending idea (cached_prs never updated). The fix recovers the
+    original branch name from the already-stored cached_prs row.
+
+    The web UX never tripped this because it leaves the branch in place
+    (the repo's default_delete_branch_after_merge is false).
+
+    The bug only manifests on an out-of-band merge (the PR merged +
+    branch deleted directly in Gitea, with the in-app merge endpoint never
+    reconciling the row while the branch still existed) -- which is exactly
+    what the #35 CLI lane does. An in-app merge reconciles cached_prs to
+    'merged' before the branch is gone, so it never trips this; the test
+    therefore drives the Gitea state directly to reproduce the CLI path.
+    """
+    from fastapi.testclient import TestClient
+    from app import db, cache, gitea as gitea_mod
+    from app.config import load_config
+
+    app, fake = app_with_fake_gitea
+
+    with TestClient(app) as client:
+        provision_user_row(user_id=2, login="alice", role="contributor")
+        sign_in_as(client, user_id=2, gitea_login="alice", display_name="Alice", role="contributor", email="alice@test")
+        r = client.post("/api/rfcs/propose", json={
+            "title": "Informed Consent",
+            "slug": "informed-consent",
+            "pitch": "A first-class definition of consent in OHM.",
+            "tags": [],
+        })
+        assert r.status_code == 200, r.text
+        pr_number = r.json()["pr_number"]
+
+        # The proposal is cached as an open idea PR.
+        items = client.get("/api/proposals").json()["items"]
+        assert any(i["pr_number"] == pr_number for i in items)
+
+        # Out-of-band CLI merge (ROADMAP #35 lane): the PR is merged AND
+        # its branch deleted directly in Gitea, WITHOUT the in-app merge
+        # endpoint ever running. So cached_prs still says state='open' and
+        # Gitea now reports the merged PR's head.ref as the sentinel. This
+        # is the exact state `rfc-authoring.sh pr-merge --delete-branch`
+        # leaves behind.
+        for pr in fake.pulls[("wiggleverse", "meta")]:
+            if pr["number"] == pr_number:
+                # land the file on main (the push side already reconciles
+                # cached_rfcs into a super-draft via the webhook/sweep)
+                for (o, rp, br, p), data in list(fake.files.items()):
+                    if (o, rp, br) == ("wiggleverse", "meta", "propose/informed-consent"):
+                        fake.files[("wiggleverse", "meta", "main", p)] = dict(data)
+                pr["state"] = "closed"
+                pr["merged"] = True
+                pr["merged_at"] = "2026-05-29T12:13:00Z"
+                pr["closed_at"] = "2026-05-29T12:13:00Z"
+                pr["merge_commit_sha"] = fake._next_sha()
+                pr["head"]["ref"] = f"refs/pull/{pr_number}/head"
+        fake.branches[("wiggleverse", "meta")].pop("propose/informed-consent", None)
+
+        # The reconcile sweep runs (a later webhook, or the 5-min safety net).
+        import asyncio
+        cfg = load_config()
+        gclient = gitea_mod.Gitea(cfg)
+        asyncio.run(cache.refresh_meta_repo(cfg, gclient))
+        asyncio.run(cache.refresh_meta_pulls(cfg, gclient))
+
+        # The bug: this used to still list informed-consent (frozen 'open'
+        # row, slug unparseable from the sentinel). The fix recovers the
+        # stored branch name, so the row reconciles to merged and the ghost
+        # is gone.
+        assert client.get("/api/proposals").json()["items"] == []
+
+        # And the cached_prs row is correctly merged, not a frozen 'open'.
+        row = db.conn().execute(
+            "SELECT state FROM cached_prs WHERE pr_number = ?", (pr_number,)
+        ).fetchone()
+        assert row["state"] == "merged", f"expected merged, got {row['state']}"
+
+        # The super-draft itself is unaffected — still in the catalog.
+        items = client.get("/api/rfcs").json()["items"]
+        assert any(i["slug"] == "informed-consent" and i["state"] == "super-draft" for i in items)
+
+
 def test_slug_uniqueness_enforced(app_with_fake_gitea):
    from fastapi.testclient import TestClient
    app, _fake = app_with_fake_gitea
@@ -94,6 +94,44 @@ def test_longest_match_wins():
    assert out[-1]["label"] == "Open Human Model"


+def test_pending_term_emits_contribute_segment():
+    # Part 3: a super-draft match is an `rfc-pending` segment carrying the
+    # owner display name, not a plain link.
+    idx = rfc_links.LinkIndex([
+        rfc_links.Term(key="open human model", kind="pending",
+                       slug="open-human-model", title="Open Human Model", owner="Alice"),
+    ])
+    out = idx.segment("see Open Human Model please")
+    assert out[1] == {
+        "type": "rfc-pending", "slug": "open-human-model",
+        "label": "Open Human Model", "title": "Open Human Model", "owner": "Alice",
+    }
+
+
+def test_candidate_term_emits_create_segment():
+    # Part 2: a candidate term carries its canonical spelling for the
+    # propose pre-fill; no slug (no RFC exists yet).
+    idx = rfc_links.LinkIndex([
+        rfc_links.Term(key="memory model", kind="candidate", term="Memory Model"),
+    ])
+    out = idx.segment("the memory model is unspecified")
+    assert out[1] == {"type": "rfc-candidate", "label": "memory model", "term": "Memory Model"}
+
+
+def test_kind_precedence_active_beats_pending_beats_candidate():
+    # All three buckets contribute the same key; the highest-precedence
+    # kind (active) must win at the position.
+    key = "open human model"
+    idx = rfc_links.LinkIndex([
+        rfc_links.Term(key=key, kind="candidate", term="Open Human Model"),
+        rfc_links.Term(key=key, kind="pending", slug="ohm-draft", title="Open Human Model", owner="A"),
+        rfc_links.Term(key=key, kind="active", slug="open-human-model", title="Open Human Model"),
+    ])
+    out = idx.segment("the Open Human Model")
+    assert out[-1]["type"] == "rfc"
+    assert out[-1]["slug"] == "open-human-model"
+
+
 def test_keys_for_gating():
    keys = lambda **kw: set(rfc_links._keys_for(**kw))
    # rfc_id always contributes.
@@ -55,13 +55,19 @@ def _outbound_otc_envelopes(to_address: str | None = None) -> list[dict]:


 def _patch_siteverify(monkeypatch, *, success: bool, error_codes: list[str] | None = None):
-    """Replace `httpx.post` inside `app.turnstile` with a stub that
+    """Replace `turnstile._siteverify_post` with an async stub that
    returns the requested success shape. The stub does not touch the
    real CloudFlare endpoint and never sees a real secret.
+
+    I4 (security-audit-0026): the siteverify call is now awaited on an
+    `httpx.AsyncClient`, isolated behind the `_siteverify_post` seam.
+    Patching that narrow function (rather than the shared
+    `httpx.AsyncClient`, which gitea/docs also construct) keeps app boot
+    intact.
    """
    captured = {}

-    def fake_post(url, *, data=None, timeout=None, **kwargs):
+    async def fake_post(url, data):
        captured["url"] = url
        captured["data"] = data
        body = {"success": bool(success)}
@@ -70,7 +76,7 @@ def _patch_siteverify(monkeypatch, *, success: bool, error_codes: list[str] | No
        return SimpleNamespace(json=lambda: body)

    from app import turnstile as turnstile_mod
-    monkeypatch.setattr(turnstile_mod.httpx, "post", fake_post)
+    monkeypatch.setattr(turnstile_mod, "_siteverify_post", fake_post)
    return captured


@@ -170,14 +176,15 @@ def test_otc_request_admits_when_secret_unset_and_not_required(app_with_fake_git

    monkeypatch.delenv("CLOUDFLARE_TURNSTILE_SECRET", raising=False)
    monkeypatch.delenv("TURNSTILE_REQUIRED", raising=False)
-    # The httpx.post inside turnstile must not be called in this path —
-    # patch it to a sentinel that explodes if it ever runs.
+    # The siteverify call inside turnstile must not be made in this path —
+    # patch the seam to a sentinel that explodes if it ever runs
+    # (I4: the seam is now `_siteverify_post`, not module-level httpx.post).
    from app import turnstile as turnstile_mod

-    def must_not_be_called(*a, **kw):
+    async def must_not_be_called(*a, **kw):
        raise AssertionError("siteverify should not run when no secret is configured")

-    monkeypatch.setattr(turnstile_mod.httpx, "post", must_not_be_called)
+    monkeypatch.setattr(turnstile_mod, "_siteverify_post", must_not_be_called)

    app, _fake = app_with_fake_gitea
    with TestClient(app) as client:
@@ -218,3 +225,27 @@ def test_otc_request_refuses_when_required_but_secret_unset(app_with_fake_gitea,
        )
        assert r.status_code == 500, r.text
        assert _outbound_otc_envelopes("alice@example.com") == []
+
+
+# ---------------------------------------------------------------------------
+# I4 (security-audit-0026): verify_token is a coroutine — calling it returns
+# an awaitable, not a VerifyOutcome. Locks the async contract so a revert to
+# the synchronous event-loop-blocking shape fails here, not just in the
+# integration paths.
+# ---------------------------------------------------------------------------
+
+
+def test_verify_token_is_async_and_soft_skips_without_secret(monkeypatch):
+    import asyncio
+
+    from app import turnstile as turnstile_mod
+
+    monkeypatch.delenv("CLOUDFLARE_TURNSTILE_SECRET", raising=False)
+    monkeypatch.delenv("TURNSTILE_REQUIRED", raising=False)
+
+    coro = turnstile_mod.verify_token("any-token")
+    assert asyncio.iscoroutine(coro), "verify_token must be a coroutine (I4)"
+    outcome = asyncio.run(coro)
+    # No secret + not required → the gate stays open without any network call.
+    assert outcome.ok is True
+    assert outcome.reason == "skipped"
@@ -18,6 +18,56 @@ server {
    listen [::]:80;
    server_name ohm.wiggleverse.org;

+    # v0.25.0 security hardening (audit 0026 M2/L8)
+    #
+    # NOTE: certbot promotes THIS server block to the HTTPS listener
+    # (`listen 443 ssl`) and adds a separate port-80 → 443 redirect
+    # block (see the install comment above). These response headers
+    # therefore ride into the HTTPS server block on the VM. They use
+    # `add_header ... always` so they also apply to nginx-generated
+    # error responses (4xx/5xx), not just 200s.
+    #
+    # `server_tokens off` (L8) — suppress the nginx version in the
+    # Server header and on error pages so we don't advertise the
+    # build to scanners.
+    server_tokens off;
+
+    add_header Strict-Transport-Security "max-age=31536000; includeSubDomains" always;
+    add_header X-Frame-Options "DENY" always;
+    add_header X-Content-Type-Options "nosniff" always;
+    add_header Referrer-Policy "strict-origin-when-cross-origin" always;
+
+    # Content-Security-Policy (M2). Tuned to what the SPA actually loads:
+    #   - default-src 'self': everything not called out below is same-origin.
+    #   - script-src 'self' + challenges.cloudflare.com: the only external
+    #     <script> tag the app injects is the CloudFlare Turnstile widget
+    #     (frontend/src/components/TurnstileWidget.jsx). Amplitude and
+    #     mermaid are BUNDLED (dynamic `import()` from node_modules, served
+    #     from 'self'), so they need no extra script origin — *.amplitude.com
+    #     is listed defensively in case a future SDK build script-injects.
+    #     script-src DELIBERATELY OMITS 'unsafe-inline' — no inline <script>
+    #     is used, so we keep XSS-via-inline-script blocked.
+    #   - style-src 'unsafe-inline' IS REQUIRED by the current build: the
+    #     JSX uses inline `style={...}` attributes throughout and mermaid
+    #     injects <style> blocks at render time. Removing it would break
+    #     layout; tightening this is a future build-side change (nonce/hash).
+    #   - img-src 'self' data: https: — markdown/RFC bodies may embed remote
+    #     images and data: URIs; svg/mermaid output uses data: too.
+    #   - font-src 'self' data: — bundled fonts plus data: webfonts.
+    #   - connect-src 'self' + *.amplitude.com + challenges.cloudflare.com:
+    #     the app's API/auth/SSE are same-origin (nginx proxy); Amplitude
+    #     Analytics + Session Replay (shipped at sampleRate 1) POST to
+    #     *.amplitude.com; Turnstile verifies via challenges.cloudflare.com.
+    #   - worker-src 'self' blob: — Amplitude Session Replay spins up a
+    #     Web Worker from a blob: URL for capture/compression; without
+    #     blob: here session replay breaks for every consenting user.
+    #   - frame-src challenges.cloudflare.com — the Turnstile challenge
+    #     renders in an iframe from that origin.
+    #   - frame-ancestors 'none' — clickjacking defense, pairs with
+    #     X-Frame-Options DENY for older agents.
+    #   - base-uri 'self'; object-src 'none' — lock down <base>/<object>.
+    add_header Content-Security-Policy "default-src 'self'; script-src 'self' https://challenges.cloudflare.com https://*.amplitude.com; style-src 'self' 'unsafe-inline'; img-src 'self' data: https:; font-src 'self' data:; connect-src 'self' https://*.amplitude.com https://challenges.cloudflare.com; worker-src 'self' blob:; frame-src https://challenges.cloudflare.com; frame-ancestors 'none'; base-uri 'self'; object-src 'none'" always;
+
    # Static SPA assets live in the Vite build output. The systemd unit
    # runs as user `rfc-app`; make sure nginx (usually `www-data`) can
    # read this path. Either group-add www-data into rfc-app's group, or
@@ -42,5 +42,32 @@ ProtectHome=true
 PrivateTmp=true
 ReadWritePaths=/opt/rfc-app/backend/data

+# v0.25.0 security hardening (audit 0026 L4) — defense-in-depth.
+# The service binds 127.0.0.1:8000 and runs plain CPython
+# (FastAPI/uvicorn + sqlite + bcrypt + httpx), so it needs no
+# capabilities and no exotic syscalls.
+CapabilityBoundingSet=
+AmbientCapabilities=
+PrivateDevices=true
+ProtectKernelTunables=true
+ProtectKernelModules=true
+ProtectKernelLogs=true
+ProtectControlGroups=true
+RestrictAddressFamilies=AF_INET AF_INET6 AF_UNIX
+RestrictNamespaces=true
+LockPersonality=true
+# MemoryDenyWriteExecute=true blocks W^X memory — safe for stock
+# CPython (no JIT) and the pure-Python/C-extension stack here, but
+# would break a JIT or a C-ext that mmaps W+X. Watch the first
+# restart's journal for a crash; if uvicorn fails to come up,
+# comment this one line out and reload.
+MemoryDenyWriteExecute=true
+RestrictRealtime=true
+RestrictSUIDSGID=true
+SystemCallFilter=@system-service
+SystemCallErrorNumber=EPERM
+SystemCallArchitectures=native
+UMask=0077
+
 [Install]
 WantedBy=multi-user.target
@@ -1,12 +1,12 @@
 {
  "name": "rfc-app-frontend",
-  "version": "0.21.0",
+  "version": "0.24.0",
  "lockfileVersion": 3,
  "requires": true,
  "packages": {
    "": {
      "name": "rfc-app-frontend",
-      "version": "0.21.0",
+      "version": "0.24.0",
      "dependencies": {
        "@amplitude/unified": "^1.1.9",
        "@codemirror/commands": "^6.10.3",
@@ -18,6 +18,7 @@
        "@tiptap/pm": "^3.5.0",
        "@tiptap/react": "^3.5.0",
        "@tiptap/starter-kit": "^3.5.0",
+        "dompurify": "^3.2.4",
        "marked": "^18.0.4",
        "mermaid": "^11.15.0",
        "react": "^19.2.6",
@@ -1,7 +1,7 @@
 {
  "name": "rfc-app-frontend",
  "private": true,
-  "version": "0.26.0",
+  "version": "0.30.2",
  "type": "module",
  "scripts": {
    "dev": "vite",
@@ -19,6 +19,7 @@
    "@tiptap/pm": "^3.5.0",
    "@tiptap/react": "^3.5.0",
    "@tiptap/starter-kit": "^3.5.0",
+    "dompurify": "^3.2.4",
    "marked": "^18.0.4",
    "mermaid": "^11.15.0",
    "react": "^19.2.6",
@@ -1365,6 +1365,30 @@
  font-weight: 500;
 }
 .rfc-autolink:hover { text-decoration-style: solid; }
+
+/* #28 Parts 2–3: a matched term that isn't a live link but carries an
+   offer (contribute to a pending RFC / create a new one). The term reads
+   as enriched (dotted underline, no link colour); the offer is a small
+   trailing chip so the prose stays readable. */
+.rfc-pending, .rfc-candidate {
+  text-decoration: underline;
+  text-decoration-style: dotted;
+  text-underline-offset: 2px;
+}
+.rfc-offer {
+  margin-left: 4px;
+  padding: 0 5px;
+  font-size: 0.74em;
+  font-weight: 600;
+  line-height: 1.5;
+  border-radius: 6px;
+  white-space: nowrap;
+  text-decoration: none;
+  border: 1px solid var(--color-border, #ccc);
+  color: var(--color-link);
+}
+.rfc-offer:hover { background: var(--color-surface-alt, rgba(0,0,0,0.04)); }
+.rfc-offer-create { border-style: dashed; }
 .pr-header-edit { display: flex; flex-direction: column; gap: 8px; }
 .pr-header-right {
  display: flex; flex-direction: column; align-items: flex-end; gap: 8px;
@@ -1,5 +1,5 @@
 import { useEffect, useRef, useState } from 'react'
-import { Routes, Route, Link, Navigate, useLocation, useNavigate } from 'react-router-dom'
+import { Routes, Route, Link, Navigate, useLocation, useNavigate, useSearchParams } from 'react-router-dom'
 import { getMe, subscribeToNotifications } from './api'
 import { anonymize, EVENTS, identify, track } from './lib/analytics'
 import { useLastState } from './lib/useLastState'
@@ -9,6 +9,7 @@ import RFCView from './components/RFCView.jsx'
 import PRView from './components/PRView.jsx'
 import ProposalView from './components/ProposalView.jsx'
 import ProposeModal from './components/ProposeModal.jsx'
+import ContributeRequestForm from './components/ContributeRequestForm.jsx'
 import Landing from './components/Landing.jsx'
 import Login from './components/Login.jsx'
 import BetaPending from './components/BetaPending.jsx'
@@ -51,6 +52,19 @@ export default function App() {
  const [identifyReady, setIdentifyReady] = useState(false)
  const navigate = useNavigate()
  const location = useLocation()
+  // #28 Parts 2–3: the LinkedText create/contribute affordances route via
+  // query params so they need no prop-threading from deep in a comment
+  // list. `?propose=<term>` opens the propose modal pre-filled;
+  // `?contribute=<slug>&term=<term>` opens the contribute-request form.
+  const [searchParams, setSearchParams] = useSearchParams()
+  const proposeParam = searchParams.get('propose')
+  const contributeSlug = searchParams.get('contribute')
+  const contributeTerm = searchParams.get('term')
+  const clearParams = (...keys) => {
+    const next = new URLSearchParams(searchParams)
+    keys.forEach(k => next.delete(k))
+    setSearchParams(next, { replace: true })
+  }
  // v0.15.0 — Page Viewed event taxonomy. We fire on every
  // route change; the analytics wrapper itself decides whether
  // anything ships out (consent + key check). The first fire is
@@ -199,7 +213,7 @@ export default function App() {
              wonders why a conversation is public can reach the answer
              in two clicks. Anonymous viewers see it too. */}
          <Link to="/philosophy" className="header-about" title="Why this exists (§14)">
-            Philosophy
+            About
          </Link>
          <Link to="/docs" className="header-about" title="User guide">
            Docs
@@ -313,17 +327,26 @@ export default function App() {
          } />
        </Routes>
      </div>
-      {proposeOpen && viewer && (
+      {(proposeOpen || proposeParam != null) && viewer && (
        <ProposeModal
          viewer={viewer}
-          onClose={() => setProposeOpen(false)}
+          initialTitle={proposeParam || ''}
+          onClose={() => { setProposeOpen(false); clearParams('propose') }}
          onSubmitted={({ pr_number }) => {
            setProposeOpen(false)
+            clearParams('propose')
            setCatalogVersion(v => v + 1)
            navigate(`/proposals/${pr_number}`)
          }}
        />
      )}
+      {contributeSlug && viewer && (
+        <ContributeRequestForm
+          slug={contributeSlug}
+          term={contributeTerm || ''}
+          onClose={() => clearParams('contribute', 'term')}
+        />
+      )}
      {inboxOpen && viewer && (
        <Inbox onClose={() => setInboxOpen(false)} lastChangeTick={inboxTick} />
      )}
@@ -226,6 +226,40 @@ export async function suggestTags({ title, pitch, useCase }) {
  }
 }

+// Roadmap #28 Part 3: offer-to-contribute-to-a-pending-RFC.
+// `contributionTarget` feeds the contribute form (RFC title, owner
+// display, the viewer's eligibility); `requestContribution` submits the
+// ask; accept/decline are the owner's inbox actions.
+export async function contributionTarget(slug) {
+  return jsonOrThrow(await fetch(`/api/rfcs/${slug}/contribution-target`))
+}
+
+export async function requestContribution(slug, { matchedTerm, whoIAm, why, useCase }) {
+  const res = await fetch(`/api/rfcs/${slug}/contribution-requests`, {
+    method: 'POST',
+    headers: { 'Content-Type': 'application/json' },
+    body: JSON.stringify({
+      matched_term: matchedTerm,
+      who_i_am: whoIAm,
+      why,
+      use_case: useCase || null,
+    }),
+  })
+  return jsonOrThrow(res)
+}
+
+export async function acceptContributionRequest(slug, requestId) {
+  return jsonOrThrow(await fetch(
+    `/api/rfcs/${slug}/contribution-requests/${requestId}/accept`, { method: 'POST' },
+  ))
+}
+
+export async function declineContributionRequest(slug, requestId) {
+  return jsonOrThrow(await fetch(
+    `/api/rfcs/${slug}/contribution-requests/${requestId}/decline`, { method: 'POST' },
+  ))
+}
+
 export async function mergeProposal(prNumber) {
  const res = await fetch(`/api/proposals/${prNumber}/merge`, { method: 'POST' })
  return jsonOrThrow(res)
@@ -0,0 +1,163 @@
+// ContributeRequestForm.jsx — roadmap #28 Part 3.
+//
+// The "ask to contribute" popover, opened from an `rfc-pending` affordance
+// in LinkedText (App reads `?contribute=<slug>&term=<term>`). It loads the
+// contribution target (RFC title + owner display + the viewer's
+// eligibility), shows the framing line "<owner> is working on an RFC for
+// '<term>'", and collects the three #15/#26-vocabulary fields:
+//
+//   * Who I am          (required, free-text)
+//   * Why I'm asking    (required, free-text)
+//   * What I'd use it for (optional, mirrors #26)
+//
+// Submitting POSTs the request, which lands in each owner's §15 inbox.
+// When the viewer isn't eligible (anonymous, already a collaborator, or
+// has a pending ask) the form shows the backend's reason instead.
+
+import { useEffect, useState } from 'react'
+import { contributionTarget, requestContribution } from '../api'
+
+export default function ContributeRequestForm({ slug, term, onClose }) {
+  const [target, setTarget] = useState(null)
+  const [loadError, setLoadError] = useState(null)
+  const [whoIAm, setWhoIAm] = useState('')
+  const [why, setWhy] = useState('')
+  const [useCase, setUseCase] = useState('')
+  const [submitting, setSubmitting] = useState(false)
+  const [error, setError] = useState(null)
+  const [done, setDone] = useState(false)
+
+  useEffect(() => {
+    let live = true
+    contributionTarget(slug)
+      .then(t => { if (live) setTarget(t) })
+      .catch(err => { if (live) setLoadError(err.message || 'Could not load this RFC.') })
+    return () => { live = false }
+  }, [slug])
+
+  async function handleSubmit(e) {
+    e.preventDefault()
+    if (!whoIAm.trim() || !why.trim()) return
+    setSubmitting(true)
+    setError(null)
+    try {
+      await requestContribution(slug, {
+        matchedTerm: term || target?.title || slug,
+        whoIAm: whoIAm.trim(),
+        why: why.trim(),
+        useCase: useCase.trim() || null,
+      })
+      setDone(true)
+    } catch (err) {
+      setError(err.message || 'Could not send your request.')
+    } finally {
+      setSubmitting(false)
+    }
+  }
+
+  const owner = target?.owner || 'The owner'
+  const label = term || target?.title || slug
+
+  return (
+    <div className="modal-overlay" onClick={e => { if (e.target === e.currentTarget) onClose() }}>
+      <div className="modal">
+        <div className="modal-header">
+          <h2>Ask to contribute</h2>
+          <button className="modal-close" onClick={onClose}>×</button>
+        </div>
+
+        {loadError && (
+          <div className="modal-body"><p className="field-error">{loadError}</p></div>
+        )}
+
+        {!loadError && done && (
+          <>
+            <div className="modal-body">
+              <p>
+                Your request has been sent to <strong>{owner}</strong>. You'll hear back
+                in your inbox; if it's accepted you'll get an invitation by email to
+                join the RFC.
+              </p>
+            </div>
+            <div className="modal-actions">
+              <button type="button" className="btn-primary" onClick={onClose}>Done</button>
+            </div>
+          </>
+        )}
+
+        {!loadError && !done && target && !target.eligible && (
+          <>
+            <div className="modal-body">
+              <p className="field-help" style={{ marginTop: 0 }}>
+                {owner} is working on an RFC for <strong>'{label}'</strong>.
+              </p>
+              <p>{target.already_requested
+                ? "You've already asked to contribute to this RFC — the owner has your request."
+                : (target.reason || 'You cannot ask to contribute to this RFC right now.')}</p>
+            </div>
+            <div className="modal-actions">
+              <button type="button" className="btn-secondary" onClick={onClose}>Close</button>
+            </div>
+          </>
+        )}
+
+        {!loadError && !done && target && target.eligible && (
+          <form onSubmit={handleSubmit}>
+            <div className="modal-body">
+              <p className="field-help" style={{ marginTop: 0 }}>
+                <strong>{owner}</strong> is working on an RFC for <strong>'{label}'</strong>.
+                Tell them a little about why you'd like to contribute.
+              </p>
+
+              <label htmlFor="contribute-who">Who I am</label>
+              <textarea
+                id="contribute-who"
+                value={whoIAm}
+                onChange={e => setWhoIAm(e.target.value)}
+                placeholder="Your name and a sentence of context."
+                rows={2}
+                autoFocus
+                required
+              />
+
+              <label htmlFor="contribute-why">Why I'm asking to contribute</label>
+              <textarea
+                id="contribute-why"
+                value={why}
+                onChange={e => setWhy(e.target.value)}
+                placeholder="What you'd bring, or what draws you to this RFC."
+                rows={3}
+                required
+              />
+
+              <label htmlFor="contribute-use-case">What I'd use the RFC for (optional)</label>
+              <textarea
+                id="contribute-use-case"
+                value={useCase}
+                onChange={e => setUseCase(e.target.value)}
+                placeholder="The concrete thing you intend to build or do with it. Optional."
+                rows={2}
+              />
+
+              {error && <p className="field-error">{error}</p>}
+            </div>
+            <div className="modal-actions">
+              <button type="button" className="btn-secondary" onClick={onClose}>Cancel</button>
+              <button
+                type="submit"
+                className="btn-primary"
+                disabled={!whoIAm.trim() || !why.trim() || submitting}
+              >
+                {submitting ? 'Sending…' : 'Send request'}
+              </button>
+            </div>
+          </form>
+        )}
+
+        {!loadError && !done && !target && (
+          <div className="modal-body"><p className="field-help">Loading…</p></div>
+        )}
+      </div>
+    </div>
+  )
+}
@@ -17,7 +17,7 @@
 import { useEditor, EditorContent, Extension } from '@tiptap/react'
 import StarterKit from '@tiptap/starter-kit'
 import { useEffect, useRef, useCallback } from 'react'
-import { marked } from 'marked'
+import { renderMarkdown } from '../lib/sanitizeHtml'
 import { Plugin, PluginKey } from 'prosemirror-state'
 import { Decoration, DecorationSet } from 'prosemirror-view'

@@ -122,7 +122,7 @@ export default function Editor({

  useEffect(() => {
    if (!editor || content == null) return
-    const html = marked.parse(content)
+    const html = renderMarkdown(content)
    editor.commands.setContent(html, false)
  }, [content, editor])

@@ -126,3 +126,13 @@
  line-height: var(--leading-normal);
  color: var(--color-text-muted);
 }
+
+/* #28 Part 3 — actionable contribute-request row: the requester's
+   who/why/use-case detail plus an Accept/Decline pair. */
+.inbox-row-action { display: flex; flex-direction: column; gap: 8px; padding: 12px; }
+.inbox-row-action .inbox-row-main { display: flex; align-items: center; gap: 8px; }
+.inbox-request-detail { margin-left: 18px; font-size: var(--text-sm); }
+.inbox-request-detail p { margin: 2px 0; color: var(--color-text-muted); }
+.inbox-request-detail strong { color: var(--color-text); }
+.inbox-request-actions { display: flex; gap: 8px; margin-left: 18px; }
+.inbox-request-outcome { margin: 0 0 0 18px; }
@@ -11,6 +11,8 @@
 import { useEffect, useMemo, useState } from 'react'
 import { Link } from 'react-router-dom'
 import {
+  acceptContributionRequest,
+  declineContributionRequest,
  listNotifications,
  markNotificationRead,
  markNotificationsReadByFilter,
@@ -164,7 +166,71 @@ export default function Inbox({ onClose, lastChangeTick }) {
  )
 }

+// #28 Part 3: the contribute-request row is the first actionable inbox
+// kind — it renders the requester's who/why/use-case inline and an
+// Accept/Decline pair that fire the owner's decision (accept reuses #12's
+// invite flow on the backend).
+function ContributionRequestRow({ item, onMarkRead }) {
+  const unread = !item.read_at
+  const x = item.extras || {}
+  const [outcome, setOutcome] = useState(null)   // 'accepted' | 'declined'
+  const [busy, setBusy] = useState(false)
+  const [error, setError] = useState(null)
+
+  async function act(accept) {
+    if (busy || outcome) return
+    setBusy(true)
+    setError(null)
+    try {
+      if (accept) await acceptContributionRequest(item.rfc_slug, x.request_id)
+      else await declineContributionRequest(item.rfc_slug, x.request_id)
+      setOutcome(accept ? 'accepted' : 'declined')
+      await onMarkRead(item)
+    } catch (err) {
+      setError(err.message || 'Action failed.')
+    } finally {
+      setBusy(false)
+    }
+  }
+
+  return (
+    <li className={`inbox-row inbox-row-action ${unread ? 'unread' : 'read'}`}>
+      <div className="inbox-row-main">
+        <span className="inbox-unread-dot" aria-hidden />
+        <span className={`inbox-cat cat-${item.category || 'unknown'}`}>{item.category || '·'}</span>
+        <span className="inbox-summary">{item.summary}</span>
+        <span className="inbox-when">{formatWhen(item.created_at)}</span>
+      </div>
+      <div className="inbox-request-detail">
+        {x.who_i_am && <p><strong>Who:</strong> {x.who_i_am}</p>}
+        {x.why && <p><strong>Why:</strong> {x.why}</p>}
+        {x.use_case && <p><strong>Use case:</strong> {x.use_case}</p>}
+      </div>
+      {error && <p className="field-error">{error}</p>}
+      {outcome ? (
+        <p className="inbox-request-outcome muted">
+          {outcome === 'accepted'
+            ? 'Accepted — an invitation has been sent.'
+            : 'Declined.'}
+        </p>
+      ) : (
+        <div className="inbox-request-actions">
+          <button type="button" className="btn-primary" disabled={busy || !x.request_id} onClick={() => act(true)}>
+            Accept
+          </button>
+          <button type="button" className="btn-secondary" disabled={busy || !x.request_id} onClick={() => act(false)}>
+            Decline
+          </button>
+        </div>
+      )}
+    </li>
+  )
+}
+
 function InboxRow({ item, onClick, onMarkRead, onClose }) {
+  if (item.event_kind === 'contribution_request_on_pending_rfc') {
+    return <ContributionRequestRow item={item} onMarkRead={onMarkRead} />
+  }
  const unread = !item.read_at
  const target = deepLink(item)
  const handle = async () => {
@@ -1,21 +1,40 @@
-// LinkedText.jsx — roadmap #28 Part 1.
+// LinkedText.jsx — roadmap #28 (Parts 1–3).
 //
-// Renders a backend-provided list of text/rfc-link segments (see
-// backend/app/rfc_links.py). RFC references in PR descriptions and
-// comments arrive pre-scanned as structured segments — this component
-// maps them onto plain text runs and anchor elements. It never renders
+// Renders a backend-provided list of text/link segments (see
+// backend/app/rfc_links.py). References in PR descriptions and comments
+// arrive pre-scanned as structured segments — this component maps them
+// onto plain text runs, anchors, and inline affordances. It never renders
 // HTML from the server (no dangerouslySetInnerHTML), so the surface is
 // XSS-safe regardless of what a comment author typed.
 //
+// Segment types:
+//   * `rfc`           — Part 1: a link to an accepted (active) RFC.
+//   * `rfc-pending`   — Part 3: the term names a pending (super-draft)
+//                       RFC; a signed-in viewer who isn't its owner gets
+//                       an inline "ask to contribute" affordance routing
+//                       to the contribute form (App reads `?contribute=`).
+//   * `rfc-candidate` — Part 2: a strong-candidate term with no RFC yet;
+//                       a viewer with create rights (`canCreate`) gets a
+//                       "create RFC" affordance routing to the propose
+//                       flow pre-filled (App reads `?propose=`).
+//
+// Affordances degrade to plain text when the viewer lacks the relevant
+// right, so the visible prose is identical for everyone — only the
+// offered actions differ.
+//
 // `segments` is the enriched array; `text` is the raw fallback used when
 // the field is absent (an older cached response, or a caller that didn't
-// pass segments). Either way the visible text is identical — only the
-// links differ.
+// pass segments).

-export default function LinkedText({ segments, text }) {
+import { Link } from 'react-router-dom'
+
+export default function LinkedText({ segments, text, viewer, canCreate }) {
  if (!Array.isArray(segments) || segments.length === 0) {
    return <>{text ?? ''}</>
  }
+  // A signed-in, beta-granted viewer can ask to contribute; the backend
+  // re-checks ownership/collaborator status and rejects self-requests.
+  const canContribute = !!viewer && viewer.permission_state === 'granted'
  return (
    <>
      {segments.map((seg, i) => {
@@ -31,6 +50,40 @@ export default function LinkedText({ segments, text }) {
            </a>
          )
        }
+        if (seg.type === 'rfc-pending') {
+          const who = seg.owner || 'Someone'
+          return (
+            <span key={i} className="rfc-pending">
+              {seg.label}
+              {canContribute && (
+                <Link
+                  className="rfc-offer rfc-offer-contribute"
+                  to={`?contribute=${encodeURIComponent(seg.slug)}&term=${encodeURIComponent(seg.label)}`}
+                  title={`${who} is working on an RFC for '${seg.label}' — ask to contribute`}
+                >
+                  ask to contribute
+                </Link>
+              )}
+            </span>
+          )
+        }
+        if (seg.type === 'rfc-candidate') {
+          const term = seg.term || seg.label
+          return (
+            <span key={i} className="rfc-candidate">
+              {seg.label}
+              {canCreate && (
+                <Link
+                  className="rfc-offer rfc-offer-create"
+                  to={`?propose=${encodeURIComponent(term)}`}
+                  title={`Create RFC for '${term}'`}
+                >
+                  + create RFC
+                </Link>
+              )}
+            </span>
+          )
+        }
        return <span key={i}>{seg.text}</span>
      })}
    </>
@@ -21,6 +21,7 @@

 import { useEffect, useRef, useState, useCallback } from 'react'
 import { Marked } from 'marked'
+import { sanitizeHtml } from '../lib/sanitizeHtml'
 import { decorateAcceptedChanges } from './trackedOverlay.js'
 import ChangeTooltip from './ChangeTooltip.jsx'

@@ -98,7 +99,7 @@ export default function MarkdownPreview({
  // synchronously with the body itself — no flash of un-decorated text.
  useEffect(() => {
    if (!hostRef.current) return
-    const html = previewMarked.parse(content || '')
+    const html = sanitizeHtml(previewMarked.parse(content || ''))
    hostRef.current.innerHTML = html
    const token = ++renderTokenRef.current
    // Reset memo so the new block set re-renders from scratch.
@@ -220,7 +220,7 @@ export default function PRView({ viewer }) {
              <h1 className="pr-title">{pr.title}</h1>
              {pr.description && (
                <p className="pr-description">
-                  <LinkedText segments={pr.description_segments} text={pr.description} />
+                  <LinkedText segments={pr.description_segments} text={pr.description} viewer={viewer} canCreate={viewer?.permission_state === 'granted'} />
                </p>
              )}
              {/* #26: the optional ground-truth use case for this change,
@@ -444,7 +444,7 @@ function PRConversation({ threads, messagesByThread, threadsByKind, seenMsgId })
              </div>
              {m.quote && <pre className="chat-msg-quote">{m.quote}</pre>}
              <div className="chat-msg-body">
-                <LinkedText segments={m.text_segments} text={m.text} />
+                <LinkedText segments={m.text_segments} text={m.text} viewer={viewer} canCreate={viewer?.permission_state === 'granted'} />
              </div>
            </li>
          )
@@ -10,7 +10,7 @@

 import { useEffect, useState } from 'react'
 import { useParams, useNavigate } from 'react-router-dom'
-import { marked } from 'marked'
+import { renderMarkdown } from '../lib/sanitizeHtml'
 import { getProposal, mergeProposal, declineProposal, withdrawProposal } from '../api'

 export default function ProposalView({ viewer, onChange }) {
@@ -161,7 +161,7 @@ export default function ProposalView({ viewer, onChange }) {
      </h3>
      <div
        className="entry-body"
-        dangerouslySetInnerHTML={{ __html: marked.parse(data.entry?.body || '') }}
+        dangerouslySetInnerHTML={{ __html: renderMarkdown(data.entry?.body || '') }}
      />

      {/* #26: the optional ground-truth use case the proposer supplied. */}
@@ -169,7 +169,7 @@ export default function ProposalView({ viewer, onChange }) {
        Intended use case
      </h3>
      {data.proposed_use_case
-        ? <div className="entry-body" dangerouslySetInnerHTML={{ __html: marked.parse(data.proposed_use_case) }} />
+        ? <div className="entry-body" dangerouslySetInnerHTML={{ __html: renderMarkdown(data.proposed_use_case) }} />
        : <p style={{ color: '#999', fontStyle: 'italic' }}>Left blank by the proposer.</p>}
    </article>
  )
@@ -28,8 +28,11 @@ function slugify(title) {
    .replace(/^-+|-+$/g, '')
 }

-export default function ProposeModal({ viewer, onClose, onSubmitted }) {
-  const [title, setTitle] = useState('')
+export default function ProposeModal({ viewer, onClose, onSubmitted, initialTitle = '' }) {
+  // #28 Part 2: a "create RFC for '<term>'" affordance pre-fills the title
+  // (App passes the `?propose=<term>` value here); the slug derives from it
+  // via the same effect that drives manual typing.
+  const [title, setTitle] = useState(initialTitle)
  const [slug, setSlug] = useState('')
  const [slugEdited, setSlugEdited] = useState(false)
  const [pitch, setPitch] = useState('')
@@ -191,7 +191,7 @@ export default function RFCDiscussionPanel({ slug, viewer }) {
          </div>
        )}
        {activeMessages.map(msg => (
-          <DiscussionMessage key={msg.id} message={msg} />
+          <DiscussionMessage key={msg.id} message={msg} viewer={viewer} />
        ))}
        <div ref={bottomRef} />
      </div>
@@ -258,7 +258,7 @@ export default function RFCDiscussionPanel({ slug, viewer }) {
  )
 }

-function DiscussionMessage({ message }) {
+function DiscussionMessage({ message, viewer }) {
  const isSystem = message.role === 'system'
  if (isSystem) {
    return (
@@ -281,7 +281,7 @@ function DiscussionMessage({ message }) {
        <div className="discussion-message-quote">"{message.quote}"</div>
      )}
      <div className="discussion-message-body">
-        <LinkedText segments={message.text_segments} text={message.text} />
+        <LinkedText segments={message.text_segments} text={message.text} viewer={viewer} canCreate={viewer?.permission_state === 'granted'} />
      </div>
    </div>
  )
@@ -0,0 +1,47 @@
+// sanitizeHtml.js — the single chokepoint for turning user-authored
+// markdown into DOM-bound HTML.
+//
+// Security audit 0026 (finding C1, Critical): every `marked.parse(...)`
+// result that reaches an `innerHTML` / `dangerouslySetInnerHTML` sink was
+// previously written raw. `marked` passes through embedded HTML and
+// `javascript:`/event-handler attributes verbatim, so any user-authored
+// document (RFC body, proposal body, proposed_use_case, transcript) was a
+// stored-XSS vector — a contributor's payload executed in the session of
+// whoever viewed it, including an admin/owner during review.
+//
+// Fix: route EVERY markdown render through `renderMarkdown` (or, for
+// already-rendered HTML, `sanitizeHtml`). DOMPurify's defaults already
+// strip <script>, on* event handlers, and javascript:/unsafe-data: URIs;
+// we add a hook so any link opening a new tab carries rel="noopener
+// noreferrer". The html profile keeps the standard markdown tag set plus
+// class + data-* attributes (the latter is what MarkdownPreview's mermaid
+// placeholder relies on); mermaid renders its SVG into the DOM *after*
+// sanitization and is itself locked down with securityLevel:'strict'.
+
+import DOMPurify from 'dompurify'
+import { marked } from 'marked'
+
+let _hookInstalled = false
+function ensureHook() {
+  if (_hookInstalled) return
+  DOMPurify.addHook('afterSanitizeAttributes', (node) => {
+    if (node.tagName === 'A' && node.getAttribute('target') === '_blank') {
+      node.setAttribute('rel', 'noopener noreferrer')
+    }
+  })
+  _hookInstalled = true
+}
+
+// Sanitize an already-rendered HTML string. Use when the HTML did not come
+// from `marked` (rare) or when a caller parses markdown with a bespoke
+// `Marked` instance and only needs the sanitize step.
+export function sanitizeHtml(html) {
+  ensureHook()
+  return DOMPurify.sanitize(html || '', { USE_PROFILES: { html: true } })
+}
+
+// Parse markdown with the shared `marked` and sanitize the result. This is
+// the drop-in replacement for `marked.parse(src)` at any HTML sink.
+export function renderMarkdown(src) {
+  return sanitizeHtml(marked.parse(src || ''))
+}
Author	SHA1	Message	Date
Ben Stull	d581010063	v0.30.2: header nav label "Philosophy" -> "About" Revert the §14.3 persistent chrome link's display text from "Philosophy" back to "About" (relabeled in v0.21.0). Display text only — route /philosophy, the header-about class, and the page are unchanged. Patch per SPEC §20.2: cosmetic, no deployment action beyond a frontend rebuild. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>	2026-05-29 06:56:14 -07:00
Ben Stull	732b23b156	v0.30.1: fix phantom pending-idea after merge-with-branch-delete refresh_meta_pulls / refresh_rfc_repo recover a PR's slug from its Gitea head.ref, which collapses to the refs/pull/<N>/head sentinel once a merged PR's branch is deleted. The slug then parsed to None, the row was skipped, and cached_prs.state froze at 'open' — so the entry showed as both a super-draft and a pending idea. Recover the real branch name from the stored cached_prs row when Gitea reports an empty or sentinel ref. Surfaced via the ROADMAP #35 operator authoring lane (CLI merge with --delete-branch); the web UX leaves branches in place so it never hit this. Regression test added; full suite 375 green. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>	2026-05-29 05:26:00 -07:00
Ben Stull	1558cc3a8b	Release v0.30.0: sync user guide (DOCS.md) with the shipped app Documentation-only minor. The guide had drifted since it was first written; brought back in sync with v0.7.0–v0.29.0: - "Signing in" rewritten: email + one-time-code, optional passcode, trust-device 30d, optional Turnstile, and the beta-request → pending → admin-grant gate, plus admin-create + invite-claim. The vestigial email allowlist is no longer described as the gate. - "Proposing a new RFC": four → five fields (optional use-case #26) + AI tag-suggestion disclosure (#27). - "Roles & permissions": documents the pending state. - New "Invitations, cross-references, and contribution requests" section (#12 owner invites; #28 auto-link / create-RFC / ask-to- contribute). - New "Privacy and cookies" section (#11/#13). No code/schema/API/config/overlay/secret change — DOCS.md is served verbatim by /api/docs. VERSION + frontend/package.json bumped to 0.30.0. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>	2026-05-29 02:12:18 -07:00
Ben Stull	8a94e26f75	Merge pull request 'Release v0.29.0: #28 Parts 2+3 — create-RFC offers + contribute-to-pending requests' (#3 ) from feature/v0.29.0-pr-rfc-create-contribute into main	2026-05-29 03:11:02 +00:00
Ben Stull	3c9109c392	v0.29.0: #28 Parts 2+3 — create-RFC offers + contribute-to-pending requests Extends the v0.26.0 (#28 Part 1) read-time scanner into three buckets in one pass — active link (Part 1), pending-RFC contribute offer (Part 3), create-RFC offer (Part 2) — precedence active > pending > candidate. The backend still emits only structured segments (never HTML), so the surface stays XSS-safe by construction. Part 2 — create-RFC offers: a multi-word tag from the #27 taxonomy with no defining RFC renders, for a create-rights viewer, as an inline "+ create RFC" affordance that opens the propose modal pre-filled (?propose=<term>; ProposeModal gained initialTitle). Conservative multi-word gate; broader heuristics + the Haiku path are deferred. Part 3 — contribute-to-pending offers: a term matching a super-draft renders, for a signed-in non-owner, an "ask to contribute" affordance with the owner's display name. It opens a 3-field request form (who/why/optional use-case); submitting lands a contribution_requests row (migration 024) and one actionable §15 notification per owner (new kind contribution_request_on_pending_rfc, personal-direct). The owner's inbox shows who/why/use-case inline with Accept/Decline. Accept fires #12's owner-invite flow with the requester as invitee and echoes a notification back; decline notifies the requester. Pre-merge idea PRs are out of scope. New endpoints: GET /api/rfcs/{slug}/contribution-target, POST /api/rfcs/{slug}/contribution-requests, .../{id}/accept, .../{id}/decline. The invite issue path was refactored into one reusable api_invitations.issue_invitation(...) chokepoint shared by the manual invite endpoint and Part 3's accept. Tests: 9 new (3 scanner-bucket unit + 6 e2e). Full suite 374 passing. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>	2026-05-28 20:10:13 -07:00
Ben Stull	019c8a9185	Merge pull request 'Release v0.28.0: security-audit-0026 I3 + I4 (HTML-email guard + async Turnstile siteverify)' (#2 ) from feature/v0.28.0-email-turnstile-async into main	2026-05-29 02:42:05 +00:00
Ben Stull	79a447c77b	Release v0.28.0: security-audit-0026 I3 + I4 (HTML-email guard + async Turnstile) Two informational findings from the Session-0026 audit, both framework-internal defense-in-depth. No operator action: no migration, no schema/config/overlay change, no deployment-facing surface. - I3: guard the dead text/html branch in email_envelope.build_envelope. No send path passes body_html; the unused branch would emit HTML built from possibly-unescaped user content (C1 stored-XSS class in the mail channel). Passing body_html now raises NotImplementedError; the arg is kept for documented future symmetry, enabling HTML mail becomes a deliberate escape-then-unguard change. - I4: make turnstile.verify_token async. The sync httpx.post ran inside the async /auth/otc/request handler, blocking the event loop up to the 10s timeout on a slow CloudFlare call. It now awaits httpx.AsyncClient via a narrow _siteverify_post seam (tests patch the seam, not the shared AsyncClient). The sole caller (main.py) now awaits it. Tests: full backend suite 365 passed. Added a coroutine-contract unit test for verify_token and flipped the email_envelope HTML test to assert the guard. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>	2026-05-28 19:12:16 -07:00
Ben Stull	fe044ed3db	Merge pull request 'Release v0.27.0: security hardening (audit 0026)' (#1 ) from feature/v0.27.0-security-hardening into main	2026-05-29 01:28:42 +00:00
Ben Stull	bd3ef269d4	Release v0.27.0: security hardening (audit 0026) Remediates the rfc-app application + deploy-config findings from the Session 0026 security audit. Cut as the "v0.25.0-security-hardening" branch (from v0.24.0); reversioned to 0.27.0 on rebase onto main since v0.26.0 (#28) shipped while this was in flight. - C1 (Critical): single sanitizeHtml.js chokepoint (DOMPurify) for every marked→innerHTML / dangerouslySetInnerHTML sink (MarkdownPreview, ProposalView x2, Editor); rel=noopener hook on target=_blank links. - H1: per-account OTC-verify lockout (migration 023, auto-applied) + per-IP throttle via new ratelimit.py; wired on otc verify/request + passcode check/verify. - M1: device_trust.lookup() single indexed-row read — cookie value is now "<row_id>.<raw_token>"; bcrypt-checks one row, not a global table scan. (Behavior change: existing device-trust cookies re-prompt once.) - M2: HTTP security headers (CSP/HSTS/XFO/XCTO/Referrer-Policy) at nginx. - M4: session cookie Secure-by-default (SESSION_COOKIE_SECURE opt-out). - M5: bounce webhook fails CLOSED (503) when secret unset, instead of open; RFC_APP_INSECURE_BOUNCE_WEBHOOK=1 dev opt-in. - L2/L3: per-IP cooldown + check-endpoint throttle. - L4: systemd sandbox knobs. L8/I1: nginx server_tokens off + TLS1.0/1.1 out. VERSION + frontend/package.json → 0.27.0; CHANGELOG documents the upgrade steps (incl. the out-of-band nginx + systemd apply, which the flotilla deploy gesture does not perform). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>	2026-05-28 18:28:10 -07:00
@@ -1 +1 @@
 .26.0
 .30.2