Reviewing code
Skill narumiruna/skills/skills/workflow-repository/reviewing-code
npx -y skills add narumiruna/skills --skill reviewing-codeAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 8 stars8 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Review diffs, pull requests, commits, patches, or source files for correctness, security, performance, maintainability, tests, and integration risk. Use for read-only review, including edge-case audits, PR preflight, and reviewer simulation; when fixes are also requested, hand confirmed failure-mode findings to hardening-code-paths when available.
SKILL.md
2.8 KB, 510 tokens by cl100k_base, as published. Nobody here has run it
Reviewing Code
Review the requested change, not the entire codebase. Treat the task as read-only unless fixes are requested.
Workflow
- Determine the exact target and comparison base from the request and repository or PR context. Use the target branch's merge base; do not assume
main. State broader file-audit scope when no diff is involved. - Infer intended behavior from the request, issue/PR context, tests, docs, and surrounding code. Label material assumptions.
- Trace changed behavior through relevant callers, contracts, state, errors, and downstream consumers. Report issues introduced, worsened, or made reachable by the change; separate directly relevant pre-existing problems.
- Check, where plausible:
- correctness, boundaries, state transitions, retries, concurrency, partial failure, and cleanup
- where repeated work without progress stops, and its worst-case time, resources, cost, and side effects
- interfaces, schemas, migrations, jobs, caches, feature flags, permissions, and configuration
- trust boundaries, authn/authz, validation, injection, secrets, logging, and sensitive data
- repeated work, I/O, queries, blocking, leaks, and expected scale
- tests for changed behavior and concrete error paths; maintainability only where it creates real cost or risk
- Run focused checks when feasible. Passing checks support but do not prove correctness.
- Stop after the relevant diff, directly affected contracts/callers, and focused evidence are covered. Distinguish confirmed findings, inferred risks, and unverified areas.
When fixes are requested, confirm the finding first, then use hardening-code-paths for bounded edge-case or failure-mode work. Do not hand off speculative or preference-only comments. Return to review the resulting diff and evidence.
Findings
Lead with confirmed findings ordered by severity:
- Critical: severe security impact, data loss, or widespread production failure.
- Major: important correctness, reliability, security, or maintainability risk that should block merge.
- Minor: real, low-risk defect worth correcting.
Each finding needs a file/line when available, concrete trigger, impact, and actionable fix. Omit nits unless requested. If no finding survives verification, say so and name meaningful residual risk or unavailable checks.
Give a merge verdict only for a PR/MR or explicit mergeability request: Approve, Approve with minor comments, Request changes, or Needs more context.
What ships with it: 1 file
319 B alongside SKILL.md
agents/
- openai.yaml319 B