| name | deep-review |
| description | Multi-round review pipeline. Spawns blind reviewers (quality, security, a11y, performance, data integrity, completeness), runs the full test suite, triages findings, fixes them, and verifies. Runs N rounds until clean. Use when the user says "deep review", "/deep-review", or wants a thorough post-implementation review.
|
| imports | ["compaction-resilience"] |
| invariants | ["0 test failures for PASS — no out-of-scope dismissals for test failures. Every failing test MUST be fixed.","In-scope/out-of-scope triage applies ONLY to code review findings, NEVER to test results.","Always use production-like config for E2E tests. No exceptions."] |
Deep Review Skill
Multi-round review pipeline with blind reviewers, full test suite runs, triage, and fix verification.
Invocation
/deep-review --rounds {N} --slug {slug} # Standard: N rounds of review
/deep-review --rounds {N} --slug {slug} --output-dir {path} # Custom output directory
References
This skill has reference files in references/:
references/review-checklist.md — Master review checklist
references/good-patterns.md — Good patterns to follow
references/bad-patterns.md — Anti-patterns to flag
Step 0: Initialize
- Read
docs/plans/{slug}/meta.md for current state
- Determine which review round to start from
- Write invariants (compaction resilience): Write this skill's
invariants: from
frontmatter to docs/plans/{slug}/invariants.md.
Round Structure
Each round follows the same pattern:
Round N:
1. Spawn blind reviewers in parallel
2. Run full test suite
3. Triage findings (in-scope vs out-of-scope)
4. Fix in-scope findings
5. Verify fixes
6. Gate check
Step 1: Spawn Blind Reviewers
Spawn reviewers in parallel — each reviewer gets the changed files WITHOUT seeing other reviewers' output:
deep-review-quality → docs/review/round-{N}/quality.md
deep-review-security → docs/review/round-{N}/security.md
deep-review-completeness → docs/review/round-{N}/completeness.md
deep-review-test-runner → docs/review/round-{N}/test-runner.md
Additional reviewers based on project type:
- If frontend:
deep-review-a11y, deep-review-react-perf (or equivalent)
- If API:
deep-review-data-integrity
- If critical system:
deep-review-chaos-monkey
Each reviewer receives:
- The list of changed files (diff against base branch)
- The plan file
- The user flow spec (if available)
- Their specific review lens
Step 2: Run Full Test Suite
CRITICAL: Run the FULL test suite, not a filtered subset.
<package-manager> run test
<package-manager> run test:e2e
Store results in the review output directory.
Step 3: Triage Findings
Classify each finding as:
In-scope fixes — Issues in changed files that should be fixed now
Out-of-scope issues — Pre-existing problems or unrelated concerns
IMPORTANT: Test failures are NEVER out-of-scope. Every failing test must be fixed.
Step 4: Fix
For each in-scope finding:
- Spawn
deep-review-fixer agent
- Apply the fix
- Run lint + affected tests
- Update finding status
Step 5: Verify
After all fixes:
- Run full test suite again
- Verify no regressions
- Check for coverage gaps
Gate Check
YES → round passes, proceed to next round
NO → treat as round failure, re-fix
Completion
After all rounds complete:
- Write
docs/review/summary.md with final verdict
- Write test results to the configured output path
- Update
meta.md with review status
# Review Summary
- **Rounds:** {N}
- **Total findings:** {count}
- **Fixed:** {count}
- **Out-of-scope:** {count}
- **Test results:** Meets Standards: {YES/NO}
- **Verdict:** {PASS/NEEDS FIXES}