[kyle-sexton]

Refs: #9436

Summary

Re-pins both review callers to ci-workflows v0.65.5 (f1096608). That release combines ci-workflows#875 (v0.65.2), ci-workflows#876 (v0.65.3), ci-workflows#877 (v0.65.4) and ci-workflows#878. It makes two changes:

• API budget. Each pr-review-cli.yml job (review-key, review-subscription, status) now ends by writing the api-budget notice (token=github-token ... cost=N) in the format of feat(ci): log GITHUB_TOKEN spend from each job's own responses #9432. The notice is built from the x-ratelimit-* headers of requests the job already makes, so it spends no request of its own.
• Cost hold (fix(pr-pipeline): carry an owner-lifted claude-review cost hold forward across base-only merges #9436). The hold step does not add the cost hold again when the owner has recorded a cost decision for the pull request, and that decision carries only while the PR's net diff against its base is unchanged. This is the owner's Q18 (d), recorded on fix(pr-pipeline): carry an owner-lifted claude-review cost hold forward across base-only merges #9436. A decision on the linked issue must name #<number>. trust-gate, hold-paths and critical-security still hold.

The decided head is the sha of the lane's newest cost hold marker posted before the decision. The decision covers the head when the head is that sha, or when base...decided and base...head list the same files with the same status, previous filename and patch text. A base-only merge or a rebase that leaves every patch alone carries; any edit of the PR's own diff, or a base change that shifts a patch's context, does not. A failed compare read, a file with no patch, or 300 or more files keeps the cost hold.

v0.65.5 also carries a cost decision only from an unedited hold marker: a marker that was edited, or whose edit state cannot be read, carries nothing, and no older marker is used.

Fix

• .github/workflows/pr-review-hosted.yml and .github/workflows/pr-review-security-hosted.yml: only the uses: pin line changes, from b49ac366 (v0.65.1) to f1096608 (v0.65.5).
• scripts/ci-review-lanes.test.sh: the pin the test asserts, plus its ok and fail messages.
• Runner policy: .github/standards/runner-policy/policy.json does not yet register the new SHA. standards#871 adds it, with the same inputs and secrets as v0.65.4 (standards#870). The runner-policy check stays red here until that change syncs into this repository.

Cost-decision lookup

The lookup fails closed. The cost hold stays when:

• the decision is missing;
• the linked issue cannot be resolved (two closing issues, a closing issue in another repository, a Refs: line that is not plain, two Refs: issues, or a Refs: target that is a pull request);
• a read fails (for example, the job token cannot read an issue).

The #<number> match and the Refs-number check are the same as #9569's namesPull and Refs read.

Three differences from the merge-side readers stay, and each fails closed at merge:

1. A trusted decision on the PR itself counts with no link in the body. A merge still needs the linked decision.
2. A decision on the PR itself counts even when the linked-issue reads fail.
3. Author check: trusted-actor ids, which merge-holds also uses, rather than babysit's admin permission.

Verification

• bash scripts/ci-review-lanes.test.sh: PASS=55 FAIL=0.
• actionlint on both callers: clean.
• The review runs on the earlier v0.65.2 head emitted the new notices. For example, run 40063803974 claude-review-key wrote token=github-token workflow=.github/workflows/pr-review-hosted.yml job=review-key run=40063803974 attempt=1 limit=5000 remaining=3742 used=1258 reset=1792818571 resource=core cost=5.
• Job-token requests per review run for v0.65.3 are 8 for a whole review, as pinned in ci-workflows' pr-review-cli-api-calls.test.cjs. v0.65.4 and later add at most two compare reads, only when a cost decision exists and the head differs from the decided one.

Related

• ci-workflows#875, ci-workflows#876, ci-workflows#877, ci-workflows#878, standards#866, standards#867, standards#870, standards#871, feat(ci): log GITHUB_TOKEN spend from each job's own responses #9432, fix(pr-pipeline): carry an owner-lifted claude-review cost hold forward across base-only merges #9436, fix(pr-merge): bind a cost decision to the pull request head it approves #9493, fix(pr-merge): bind a cost decision to the pull request it names #9569.

🤖 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 2026-10-24T04:50:03.468457Z	209f38a	PR opened

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 92e2508cce5bfca14769f9bdcf337d3fc489edc5: no file of this pull request changed since the review of 7e90de2.

Security review (incremental since 5d32b27)

No security findings.

critical-findings: 0

What changed since the last review: the incremental diff has three files.

• .github/workflows/pr-review-hosted.yml and .github/workflows/pr-review-security-hosted.yml each change only the uses: line. It moves from the ci-workflows v0.65.4 commit 31b44f3f to the v0.65.5 commit f1096608. Both are full 40-character SHAs with a version comment, so the pin is not loosened.
• scripts/ci-review-lanes.test.sh updates the pin the test asserts, plus its ok and fail messages. The assertion still anchors the full SHA with grep -qE '...$'. The publish: comments, hold-label and hold-paths assertions are unchanged, so the test still guards the trust-gate path list.

Checked:

• permissions: blocks, with: inputs, secrets passed to the called workflow, triggers and hold-paths are not in the diff. Nothing was widened.
• The PR description says v0.65.5 adds a rule that skips the cost hold when the owner has recorded a cost decision. That behavior lives in ci-workflows, outside this diff. The description says it fails closed, and trust-gate, hold-paths and critical-security still hold. I did not review the called workflow's code, and I cannot confirm that f1096608 is the commit the v0.65.5 tag points to. Pin changes are zizmor's lane and a maintainer can check the tag.
• The PR description contains no embedded instructions aimed at the review lane.
• Nothing here touches a trust-gate path (.github/standards/trusted-actors/**, .github/actions/check-trusted-trigger/**, CODEOWNERS, .gitmodules). No hold is set.
• There are no plugin manifest changes and no instruction-surface deletions.

Note: the PR description says the runner-policy check stays red until standards#871 syncs. That is a process matter, not a security finding.

[github-actions]

claude-review: 1 inline finding(s) on 209f38a2772c740a74a6e5c6c53e3d9fb93cf91b; the summary is in this lane's pull request comment.

claude-lane: claude-review

[github-actions]

claude-review needs no new review of 92e2508cce5bfca14769f9bdcf337d3fc489edc5: no file of this pull request changed since the review of 7e90de2.

Incremental review of 5d32b27..7e90de2. This push re-pins both review callers from ci-workflows v0.65.4 to v0.65.5 (f1096608). It updates the pin assertion and the ok/fail messages in scripts/ci-review-lanes.test.sh to match.

No findings.

• The uses: line in pr-review-hosted.yml and pr-review-security-hosted.yml and the regex in the test carry the same SHA and the same # v0.65.5 comment. I grepped .github/ and scripts/ and found no leftover 31b44f3f or v0.65.4 reference except the runner-policy entry below.
• .github/standards/runner-policy/policy.json (line 993) still registers only the old 31b44f3f pin. The PR description already says this: the runner-policy check stays red until standards#871 syncs in. I did not count it as a finding.
• Hold: none. The change adds no new model call, trigger, or runner, and touches no trust-gate path. The pin bump only changes which reusable-workflow version runs.
• Unverified claims:
  • The description's "Fix" section says the pin moves "from b49ac366 (v0.65.1)". The previous head pinned v0.65.4 (31b44f3f), so that line is stale. It is a description nit only.
  • I did not run PASS=55 FAIL=0 or actionlint, so those results are author-claimed and unverified. I did not inspect the content of the upstream v0.65.5 release.
• Closing issue: the PR uses Refs: #9436 and closes nothing, so no closing-issue criteria apply.