[kyle-sexton]

Refs: melodic-software/ci-workflows#880

Summary

Re-pins both review callers to ci-workflows v0.65.7 (4c531cea137f5cbb38d4c132dbfa6e38f76fba7d), which carries owner decisions Q52 and Q53 (2026-10-24). A trusted-path match no longer holds a pull request by itself: hold-paths now names files the prompt sends to extra security scrutiny. A cost hold now needs a real charge. A spend from prepaid credits or subscription usage becomes a summary line, not a hold.

Fix

• pr-review-hosted.yml and pr-review-security-hosted.yml: pr-review-cli.yml@4c531cea… # v0.65.7. The hold-paths comment now matches the new behavior. The input and its four globs are unchanged.
• scripts/ci-review-lanes.test.sh: pins v0.65.7, as chore(ci): re-pin the review callers to ci-workflows v0.65.6 #9660 did for v0.65.6.
• The runner-policy contract for v0.65.7 arrived through standards#876 and its sync, chore: sync standards components #9688.

Verification

• bash scripts/ci-review-lanes.test.sh: PASS=55 FAIL=0.
• node .github/standards/runner-policy/runner-policy.mjs --root <worktree>: Runner policy passed.
• actionlint passes on both callers.
• scripts/check-affected.sh --tests: 10 passed, 4 skipped (CI only).

Related

• melodic-software/ci-workflows#880
• chore(ci): re-pin the review callers to ci-workflows v0.65.6 #9660 (the v0.65.6 re-pin)
• Follow-up, not in this PR: the plugins/review code-review and security-review skills still list trust-gate and the broader cost class under "Hold for a person". The v0.65.7 wrapper prompt overrides them, and their recheck trigger has fired.

🤖 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 2026-10-24T16:52:25.473901Z	9c8f556	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-review reviewed 9c8f5561eb723ea1b08f855b2352f7153b5a426f: 0 finding(s), 0 as inline comments.

No findings. I read the whole 3-file diff and the PR context, and did not run the tests or the runner-policy check myself.

• Pin consistency: pr-review-hosted.yml and pr-review-security-hosted.yml both move to pr-review-cli.yml@4c531cea… # v0.65.7. scripts/ci-review-lanes.test.sh expects the same SHA and tag in its grep and in its pass and fail messages. No v0.65.6 text is left in the repo apart from the old, still-allowed policy entry.
• Runner policy: .github/standards/runner-policy/policy.json already has an allowlist entry for pr-review-cli.yml@4c531cea… (line 1056), so the new pin will pass the runner-policy check.
• hold-paths comments: The new comment matches the PR description: a hold needs a real-charge cost or a CRITICAL security finding, and hold-paths only sends files to extra scrutiny. The hold-paths lists are unchanged and still equal in both callers, and the test script's list check still applies.
• Verification claims: The PR's test results (PASS=55, actionlint, runner-policy) are author-claimed and unverified by me. I could not confirm that the SHA matches the v0.65.7 tag, because I have no GitHub access.
• Cost: This change only re-pins the review workflow version. It adds no new model call, trigger, schedule or paid runner, so there is no hold.
• Manifests, closing issues, paved paths: The diff touches no plugin manifests and the PR closes no issues. It only references ci-workflows#880. I did not look for a paved-path row, because the diff adds no second way to do anything.
• Left undone: The follow-up the PR names, plugins/review skills still listing trust-gate and the broader cost class, is out of scope for this PR.

[github-actions]

claude-security-review reviewed 9c8f5561eb723ea1b08f855b2352f7153b5a426f: 0 finding(s), 0 as inline comments.

Security review of PR #9803: no security findings (0).

I read the whole diff and the PR context. The PR re-pins two reusable-workflow callers from ci-workflows v0.65.6 to v0.65.7. It also updates the pin check in scripts/ci-review-lanes.test.sh and the hold-paths comments.

• Pin change: both callers still use a full 40-character commit SHA with a version comment (pr-review-hosted.yml:169, pr-review-security-hosted.yml:174). The pin is not loosened. I could not confirm that 4c531cea… is the v0.65.7 tag, because I have no network or GitHub access. The PR says it is, and the test script checks that the SHA and the version comment agree. A reviewer can compare the SHA against the release tag if they want that checked.
• Permissions and inputs: permissions: is unchanged (contents: read, pull-requests: write). So are hold-label: needs-human, publish: comments, and the four hold-paths globs. The test script still asserts all of these. No trigger changed, and no expression from the github context was added to any run: block.
• Behavior change, not a finding: a match on hold-paths no longer holds a PR on its own. The comments now say those paths only send the prompt to extra security scrutiny. A cost hold now needs a real charge. A CRITICAL security finding still holds. The PR states these are owner decisions Q52 and Q53 from 2026-10-24, so I took the narrowing as intended. It does mean a change to .github/standards/trusted-actors/**, check-trusted-trigger/**, CODEOWNERS or .gitmodules is no longer held for a person by this lane. Any protection for those paths now comes from CODEOWNERS review and branch protection, which I did not examine.
• Instruction surface: the diff changes only YAML comments and the test script's pin. It deletes no standing instruction. The follow-up the PR names is that the plugins/review skills' "Hold for a person" text is out of date. I applied the wrapper's hold classes, which replace that text.
• Protected-class list: I did not read docs/conventions/instruction-exception-register/README.md. The diff removes no instruction text, so I judged it unnecessary.
• Plugin manifests: none changed, so there are no manifest <field> lines.
• Cost: no paid plan, billing or paid-resource change. More or fewer review runs would draw on prepaid credits or subscription usage, and this change does not alter that.

critical-findings: 0

[github-actions]

The pr-verify verifier failed head 9c8f5561eb723ea1b08f855b2352f7153b5a426f. What it wrote, as data:

Summary: PR #9803 has no linked issue, so the spec is its own body. It re-pins both review callers from ci-workflows v0.65.6 to v0.65.7 and updates the test pin. The edits match the body and the test still checks them. The problem is what v0.65.7 does: a change to trusted-actors, CODEOWNERS, .gitmodules or check-trusted-trigger no longer holds the PR by itself, and spend from credits or subscription usage no longer triggers a cost hold. Both are holds the owner keeps, so loosening them is a blocking defect. The body cites owner decisions Q52 and Q53, but no trusted text available here records them.
- The PR body says the plugins/review code-review and security-review skills still list trust-gate and the broader cost class under 'Hold for a person' and leaves that as follow-up. That mismatch between the skill docs and the wrapper prompt should be tracked in an issue.
- `.github/workflows/pr-review-hosted.yml:169`: the re-pin to 4c531cea (v0.65.7) adopts upstream behavior under which a hold-paths match never holds by itself. This removes the automatic trust-gate hold on takeover-set paths (`.github/standards/trusted-actors/**`, `.github/actions/check-trusted-trigger/**`, `**/CODEOWNERS`, `**/.gitmodules`), leaving only whatever the model review decides. That loosens a hold the owner keeps. Fix: link an owner-written issue that records decisions Q52/Q53, or keep the v0.65.6 pin until one exists.
- `.github/workflows/pr-review-hosted.yml:192`: the rewritten hold-paths comment correctly describes the v0.65.7 behavior, and the input and its four globs are unchanged in both callers.

[github-actions]

The pr-verify verifier failed head 9c8f5561eb723ea1b08f855b2352f7153b5a426f. What it wrote about this file, as data:

- `.github/workflows/pr-review-security-hosted.yml:174`: the same re-pin narrows the cost hold to real charges, so prepaid-credit and subscription spend becomes a summary line instead of a hold. Cost holds belong to the owner. This needs owner authorization in trusted spec text, not only in the PR body's citation.

[github-actions]

The pr-verify verifier failed head 9c8f5561eb723ea1b08f855b2352f7153b5a426f. What it wrote about this file, as data:

- `scripts/ci-review-lanes.test.sh:79`: the pin assertion is updated to the v0.65.7 SHA and would fail if a caller drifted. Nothing here is wrong; it would need reverting only if the pin is reverted.