[kyle-sexton]

Refs: #9276

Summary

PR 7 of the native-merge plan. The review lanes in ci-workflows v0.64.0 pr-review-cli.yml accept an optional hold field (criteria from cost, trust-gate, critical-security, plus a reason), and the prompt tells the model to return one "only when the review command says to hold the pull request for a person". Until now neither review skill said when to do that. This PR adds that rule.

Fix

• /review:code-review: return hold only for cost (a change that adds or raises model or API spend, the class /source-control:babysit-prs rates cost-review-required), trust-gate (.github/standards/trusted-actors/**, .github/actions/check-trusted-trigger/**, CODEOWNERS, .gitmodules), and critical-security only where no security lane runs. A change in a hold class is no longer eligible for the trivial-change skip.
• /review:security-review: return hold only for critical-security (CRITICAL only, per owner answer OD-4) and trust-gate.
• Both skills: a hold is not a finding and adds nothing to the findings or critical-findings count. Every other flag, manifest flag lines included, stays a normal finding. With no hold field in the schema (older wrappers, seat-run mode), the skills return none. The skills write no Hold: lines.
• Changelog fragment .changes/review/, bump: minor.

Verification

• /skill-quality:check on both skills: PASS, 0 errors. The warnings were already there (body length, no ## Next) apart from the expected body-growth warning.
• scripts/check-affected.sh --tests: 12 passed, 2 skipped (no shell, no Python changed).
• The canary that checks a hold returned by the model runs with PR 6 / PR 8, as the plan describes.

Related

• Plan item 7; feat(pr-review-cli): add an optional hold label step ci-workflows#864 (the apply-review-hold composite that reads the field).
• The hold-paths input applies trust-gate by path whatever the model returns.

🤖 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 2026-10-23T05:25:57.333396Z	9f75cdb	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 reviewed a0a06fde3b0786aad5831a273aa18250cf55b323: 0 finding(s), 0 as inline comments.

No security issues found in this pull request at a0a06fd.

I read the full diff. It touches three files: a changelog fragment, plugins/review/skills/code-review/SKILL.md and plugins/review/skills/security-review/SKILL.md.

• Instruction-surface lens: The diff deletes, narrows or softens no standing instruction. It adds a hold rule and tightens two skip gates. A trust-gate change is no longer eligible for the security lane's no-surface skip, and a change in a hold class is no longer eligible for the code lane's trivial-change skip. Both gate changes make review more likely, not less. I did not read docs/conventions/instruction-exception-register/README.md; no rule was removed, so there was nothing to check against it.
• Hold escalation: The new hold field only routes a pull request to a person. A model that is manipulated into omitting it falls back to the existing path-based trust-gate enforcement and the critical-findings gate. The skills keep CRITICAL findings counted and carrying their severity: CRITICAL line, so the hold adds no bypass.
• Embedded instructions: The skill text adds no new way for the pull request body or context to instruct the lane. The existing DATA framing is unchanged.
• Manifests: No plugin.json or marketplace.json file changed, so there are no manifest <field> lines.

critical-findings: 0

[github-actions]

claude-review reviewed a0a06fde3b0786aad5831a273aa18250cf55b323: 0 finding(s), 0 as inline comments.

No findings.

I read the whole diff (3 files) and the PR context. Neither the PR body nor the issue context asks for anything beyond what the diff does.

• plugins/review/skills/code-review/SKILL.md: The skip-gate change and the new "Hold for a person" section agree with each other. critical-security is limited to repositories with no security lane, which matches the Gotchas section. The section says a hold adds nothing to the findings count and that manifest <field>: flag lines stay normal findings.
• plugins/review/skills/security-review/SKILL.md: The trust-gate carve-out in skip-gate item 2 matches the new Reporting section. The hold adds to neither the severity: CRITICAL line nor critical-findings: <n>.
• Volatile-specific rule: Both pointers to the ci-workflows hold schema carry an as-of date and a recheck trigger, as .claude/rules/skill-bodies-state-current-rules.md requires.
• Changelog fragment: It is present under .changes/review/ with bump: minor. A new optional output behavior fits that level.

I did not fetch the pinned ci-workflows v0.64.0 schema, so I could not confirm that the hold shape and the cost, trust-gate and critical-security names match it. That rests on the PR description (author-claimed). I also did not run the skill-quality check or the affected-tests run that the PR's Verification section reports, so those results are unverified by me.