| name | review-gate |
| description | Run an independent semantic review gate before merging non-trivial work, on top of (not duplicating) Alibaba Code Review and CI. Use after implementation and verification, especially for medium/high-risk changes. |
Review Gate
Use this after implementation and before merge.
Division of Labor
- Mechanical checks — owned by Alibaba Code Review and CI: linting, formatting, conventional style, and straightforward static-analysis findings. Alibaba Code Review's output is recorded via
ai-agent-pr-metadata.
- Semantic review — owned by this gate: requirements compliance, correctness, regressions, security/authorization, contract/integration risk, and acceptance-test adequacy — everything mechanical tooling can't judge.
Workflow
- Confirm linked issue and acceptance criteria.
- Confirm verification commands and results.
- Read Alibaba Code Review's findings and the GitHub CI/check results on the PR before starting the independent review — don't re-derive what's already there.
- Review spec compliance before code style.
- Check business logic and edge cases, regressions and compatibility, security and authorization, API/database/queue/integration contracts, and test adequacy including missing acceptance tests.
- Revisit an Alibaba Code Review or CI finding only when it indicates an unresolved correctness, security, data-loss, configuration, or acceptance-criteria issue — not to relitigate style or formatting.
- Post concrete findings or explicitly state no blocking issues. Do not repeat resolved lint, formatting, conventional-style, or straightforward static-analysis findings Alibaba Code Review or CI already covered.
- Apply review findings before merge.
Reviewer Contract
The reviewer should read the issue, acceptance criteria, Alibaba Code Review's output, CI results, and the PR diff first. Pull more context only when needed.
Merge Rule
CI remains a hard merge gate; this review does not replace it and is not the place to rerun or restate lint/static-analysis automation. Do not merge non-trivial work without a review record. If an automated or subagent reviewer stalls, post a manual expert review that states residual risk.