[kyle-sexton]

Refs: melodic-software/claude-code-plugins#9589

Summary

Registers pr-review-cli.yml@31b44f3ffc7f208fb281dd2b64decce67c53d316,
which is ci-workflows
v0.65.4.
claude-code-plugins re-pins its review callers to it in
claude-code-plugins#9589.

Fix

v0.65.4
(ci-workflows#877)
carries a recorded cost decision only while the pull request's net diff
against its base is unchanged. In pr-review-cli.yml the only non-comment
change is BASE_REF added to the hold step's env. The workflow_call
inputs differ only in one input description, secrets and permissions are
untouched, and the runner is unchanged. The contract copies the v0.65.3
entry from #867
forward verbatim, with a copy-forward test.

Verification

• node components/runner-policy/runner-policy.mjs --root .: Runner policy
  passed.
• node --test components/runner-policy/runner-policy.test.mjs: 346 pass,
  0 fail, including the new copy-forward test.

Related

• #867 (v0.65.3)
  and #866
  (v0.65.2)

🤖 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 2026-10-24T06:38:07.799589Z	4c5a326	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 4c5a326318d9f229841926928751c10390ebcdd7: 0 finding(s), 0 as inline comments.

No findings.

What I checked

• I compared the new pr-review-cli.yml@31b44f3ffc7f208fb281dd2b64decce67c53d316 entry in components/runner-policy/policy.json with the v0.65.3 entry at f133c6651f866845e18595d26cf59d4043b5b642 (policy.json:972-992). Routing (hosted-only), fixedRunsOn, the 11 allowedInputs and both allowedSecrets match field for field.
• The new copy-forward test asserts exactly that equality. It also asserts the new SHA appears once in the contract map, which guards against a duplicate key.
• The new entry sits next to the other pr-review-cli.yml entries. It is valid JSON in the diff and adds no new input or secret.

What I could not check

• I could not run node --test or runner-policy.mjs here, because the sandbox required approval for those commands. The PR's claims of 346 passing tests and "Runner policy passed" are author-claimed and I have not confirmed them.
• I did not read the upstream ci-workflows v0.65.4 diff. The claim that only BASE_REF was added to a step env, with inputs, secrets and runner unchanged, is also author-claimed. The registration is correct if that claim holds.

Hold classes: none. This is a policy registration with no new model spend, no trusted-actor, CODEOWNERS or .gitmodules change, and no security finding.

PR body: it opens with Refs: #… and has the four required sections.

[github-actions]

claude-security-review reviewed 4c5a326318d9f229841926928751c10390ebcdd7: 0 finding(s), 0 as inline comments.

Security review: PR #870 (head 4c5a326)

No security findings.

critical-findings: 0

Scope. The PR changes two files: components/runner-policy/policy.json and its test, components/runner-policy/runner-policy.test.mjs. It adds one allowlist entry for pr-review-cli.yml@31b44f3ffc7f208fb281dd2b64decce67c53d316, which the commit message gives as ci-workflows v0.65.4.

Checks.

• Copy-forward. The new entry matches the existing v0.65.3 entry (@f133c665…) at policy.json lines 972–992. It has the same hosted-only routing, fixedRunsOn of ubuntu-24.04, the same 11 allowedInputs, and the same two allowedSecrets mappings (REVIEWS_CLAUDE_API_KEY and REVIEWS_CLAUDE_CODE_OAUTH_TOKEN). The diff adds no new input, secret, permission, or self-hosted routing.
• Pin. The reference is a full 40-character commit SHA, not a mutable tag or branch.
• Test. The new test asserts the entry deep-equals the v0.65.3 entry and that exactly one key ends in the new SHA. It adds no weakening of an existing assertion.
• Trust-gate paths. The PR touches none of .github/standards/trusted-actors/**, .github/actions/check-trusted-trigger/**, CODEOWNERS, or .gitmodules. No hold is needed.
• Instruction surfaces and plugin manifests. The PR touches neither.

Not verified. I could not confirm that the SHA is the v0.65.4 tag, or that the release changed only BASE_REF in the hold step env. I did not run the node comparison or the test (the sandbox required approval), and I did not fetch the upstream workflow. This registration is the policy file's intended purpose, and a maintainer should confirm the pin when merging.