[kyle-sexton]

No related issue: this clears the false positive that post-merge record #11658 reports; that record is bot-written with no body, so it holds no spec to verify against

Summary

The silent-revert canary read #11646's widening of pr-verify's chunk slots as a 401-line removal of #11548, and post-merge opened #11658, so the unattended merge path sits at rung off. Nothing was lost, so this records the review in scripts/silent-revert-acknowledged.txt.

This PR's title and body are its spec. Done when: the row is on main, and the next silent-revert canary run on main passes over 5cce729. #11658 is then closed directly, as its body says, not by this PR.

Fix

One row for 5cce729c61018f709e9fa5216e639112e7f82daa, the form the canary's own output asks for.

Verification

git diff --unified=0 fd862a49b 5cce729c6 over the four files the finding names, read line by line:

• Every removed line in pr-verify.yml, workflow.test.mjs and write-verdict.mjs is a 4-slot form replaced by its 12-slot one: step chains, MAX_CHUNKS: "4", timeout-minutes: 125, ^[1-9]$, hard-coded SLOTS.
• id: judge steps in pr-verify.yml went from 15 to 39 (3 jobs of 13 steps), and all 3 judge-spec steps remain.
• fix(pr-pipeline): verify a diff of up to 12 chunks instead of stopping at 4 #11646's own changes, git show 5cce729c6, line by line:
  • docs/conventions/pr-pipeline/README.md: 4 lines removed, 4 added. The removed lines are the "at most 4 chunks" and "more than 4 chunks (over about 1.2 MB)" sentences; the added lines say "up to 12" and "more than 12 chunks (over about 3.6 MB)". Same rule, new limit.
  • workflow.test.mjs and write-verdict.mjs: 14 lines removed. Each is a 4-slot form (SLOTS = [1, 2, 3, 4], ^chunk-[1-9]$, ^[1-9]$, a 4-step chain) whose 12-slot form is added in the same commit.
  • pr-verify.yml: the removed lines repeat 2 or 3 times, once per judging job (key, subscription, fallback): timeout-minutes: 125, MAX_CHUNKS: "4", ^[1-9]$, the 4-slot if:, handoff: and leg-clean: chains, and the comment saying five 22-minute steps fit in 125 minutes. Each has a 12-slot or 301-minute replacement in the same commit.

The canary's count is line attribution across a regenerated block, not lost content. Intentional-removal: in #11646's body would have avoided the fire.

Related

• Post-merge canary failure after #11627: unattended merging demoted #11658: the demotion record.
• fix(pr-pipeline): verify a diff of up to 12 chunks instead of stopping at 4 #11646, fix(pr-pipeline): verify a large diff in chunks with a final spec judge #11548: the two merges the finding names.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-31T18:48:52.199690Z	0e9b674	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[github-actions]

claude-security-review needs no new review of b8fb933556c77fdd981d00d442906411117a5843: no file of this pull request changed since the review of 13eb60f.

No security findings.

The change adds one row to scripts/silent-revert-acknowledged.txt, for the full 40-character sha 5cce729c61018f709e9fa5216e639112e7f82daa (#11646). The file matches none of the TRUSTED PATHS globs.

• Who reaches it: the row clears only that one commit. scripts/check-silent-revert.sh reads the file at line 358 and matches rows on the full sha (lines 470-486), so it cannot cover any other commit. A malformed or wrong row fails closed, because the commit then fires the canary.
• What it gets: it only stops the canary from flagging fix(pr-pipeline): verify a diff of up to 12 chunks instead of stopping at 4 #11646's 4-to-12 chunk-slot widening in pr-verify.yml, workflow.test.mjs and write-verdict.mjs. It changes no token scope, trusted-actor list, kill switch or merge hold, and it adds no code or workflow.
• Limit on this review: I had no git access, so I did not re-check the PR's claim that the removed lines are all 4-slot forms replaced by 12-slot ones. That claim is reviewable on fix(pr-pipeline): verify a diff of up to 12 chunks instead of stopping at 4 #11646, where the judge-spec steps are said to remain. The row follows the file's audit-trail format and is not a mute list.
• Embedded text: the PR description contains no instructions aimed at the review lane.
• Instruction-surface check: no standing instruction is deleted or narrowed.
• Manifests: no plugin manifests are touched.
• Spend: no billing or prepaid-credit effect.

critical-findings: 0

[github-actions]

claude-review needs no new review of b8fb933556c77fdd981d00d442906411117a5843: no file of this pull request changed since the review of 13eb60f.

No findings. The PR adds one row to scripts/silent-revert-acknowledged.txt for 5cce729c61018f709e9fa5216e639112e7f82daa, which acknowledges #11646's widening of #11548's chunk slots.

What I checked myself, by reading the file:

• The row follows the file's documented format <full-40-char-sha> <note>. The sha is 40 hex characters, and the note is on the same line as the sha.
• The row sits with the other single-line entries, and the comment header and earlier rows are untouched.
• The note records why the removal was intended, which is what the file's audit-trail rule asks for.

What I could not check: I have no shell or git access in this run. I did not confirm that 5cce729c… is the commit the canary flagged. I also did not re-run the git diff fd862a49b 5cce729c6 comparison. The PR body's claims about that diff are the author's and are unverified by me: the 15 to 39 id: judge step count, all 3 judge-spec steps remaining, and the 4-slot to 12-slot replacements. The PR body says the author did not separate the 3 README lines attributed to #11646 from other PRs' edits. That is a stated gap in the author's review. It is not a defect in this change.

Embedded text: the PR context file contained no directives aimed at the lane. The scoped criteria are met: PR #11669 only refs #11658 and does not close it, so there are no closing-issue criteria to check.

Trusted paths: no changed file matches a TRUSTED PATHS glob. The change is a data-only allowlist row for a CI canary. It carries no cost or critical-security risk, so I return no hold.

[github-actions]

The pr-verify verifier failed head 0e9b674d33f9b241bc78ffb85acd77945fa9951e. What it wrote, as data:

Summary: Fail: there is no trusted spec. The PR's only link is `Refs: #11658`, and that issue's body
  is null in spec-text.json because a bot (github-actions) wrote it and it has no trusted comments
  or refinement. With no issue carrying trusted spec text, the rules require a fail. The diff itself
  adds one row to scripts/silent-revert-acknowledged.txt that exempts commit 5cce729c6 (#11646) from
  the silent-revert canary.
- spec-text.json: #11658 is the only linked issue, and its `body` is null with no comments. A bot
  (github-actions) wrote it, so the PR cannot be checked against any trusted requirement. To fix:
  have a trusted person write or edit #11658's body (or file a human-written issue) stating what to
  acknowledge and when it counts as done, then link that issue from the PR.
- scripts/silent-revert-acknowledged.txt:43: this row stops the silent-revert canary from flagging
  5cce729c6, which turns off a regression guard for that commit. Only the PR body supports the claim
  that #11646 lost nothing (its own `git diff` reading), and the author says they did not check the
  README lines one by one. A person should confirm that before this exemption is accepted.
- upstream-jobs.json: select-tests ran and every test suite was skipped. That is consistent with a
  data-only change to this .txt file, so the skips are not a defect.

This pull request leaves the merge queue, and ci-status holds it out while it carries pr-verify-failed. A pr-verify pass at a new commit removes the label.