| name | review |
| description | Repository review entrypoint for requests like レビューして, 監査して, 問題点を見て, 改善点を洗い出して. Runs parallel code, performance/security, and spec-consistency review, then consolidates actionable findings. |
| user_invocable | true |
Workflow
実装を進めること自体が主目的なら dev を使い、この skill は評価・監査・改善点抽出が主目的の時に使う。
Step 1: Launch 3 review agents in parallel
CRITICAL: All 3 agents MUST be launched in a SINGLE message using multiple Agent tool calls. Do NOT launch them sequentially. Send one message containing all three Agent invocations at once.
Launch the following agents in parallel:
-
code-reviewer agent
- Focus: code quality, test conventions, runtime regressions, boundary violations
- Check test-conventions compliance (no
.skip, no .only, naming, E2E coverage)
- Verify docs are in sync with implementation
-
perf-reviewer agent
- Focus: OWASP Top 10, N+1 queries, resource leaks, performance
- Check for injection, XSS, auth bypass, sensitive data exposure
- Check for unbounded queries, memory leaks, blocking I/O, bundle size
-
spec-reviewer agent
- Focus: specification consistency, over-engineering detection
- Compare implementation against plan in
work/plans/active/
- Flag premature abstractions, unnecessary complexity, scope creep
- Flag missing boundary validation, insufficient error messages
Step 2: Collect and consolidate results
After all 3 agents return:
- Gather all findings from each agent
- Deduplicate overlapping findings (keep the more detailed version)
- Sort by severity: Critical > High > Medium > Low
Step 3: Gate on Critical and High findings
If any Critical or High findings exist:
- List each finding with its agent source, file, line, and description
- State clearly: "Critical/High findings must be fixed before proceeding"
- Do NOT mark the review as passed
If no Critical or High findings exist:
- Summarize Medium and Low findings as recommendations
- Mark the review as passed
Step 4: Write review results
Write the consolidated review report to work/reviews/active/ with the following structure:
## Review Report — <date>
### Status: PASS | FAIL
### Critical / High (must fix)
- [agent] severity | file:line | description | fix suggestion
### Medium / Low (recommended)
- [agent] severity | file:line | description | fix suggestion
### Summary
- Total findings: N
- Critical: N, High: N, Medium: N, Low: N
- Agents: code-reviewer, perf-reviewer, spec-reviewer