Check Your Work
Orchestrate specialized agents to detect bugs, security issues, and quality problems in code changes.
<when_to_use>
When to Use
Invoke when user says:
- "check your work"
- "review what we wrote"
- "quality check the new code"
- "check for bugs in [files/feature]"
- After completing feature implementation
- Before committing significant changes </when_to_use>
| Phase | Agents | Action |
|---|---|---|
| 1 | - | Scope Discovery (identify files) |
| 2 | 6 parallel | All Checks (duplicate-detector, jenny, deep-bug-hunter, 3 reviewers) |
| 3 | up to 3 | Severity Validation (one agent per P0/P1/P2 level with findings) |
| 4 | 1 conditional | Deep Investigation (root cause for validated P0) |
| 5 | - | Consolidated Report + User Decision |
For Phase 2 details: references/phase-2-quality-checks.md For Phase 3 details: references/phase-3-validation.md For report format: references/consolidated-report.md
Phase 2 (6 Parallel)
| Agent | Focus |
|---|---|
| duplicate-code-detector | Reimplemented utilities/services |
| jenny | User voice - did we build what was requested? |
| deep-bug-hunter | Logic errors, race conditions, edge cases |
| Security Reviewer | SQL injection, org_id isolation, XSS, PII |
| Performance Reviewer | Memory leaks, bug-causing antipatterns |
| Correctness Reviewer | Logic errors, null handling, async issues |
Phase 3 (up to 3 Parallel)
| Agent | Focus |
|---|---|
| p0-validator | Validates P0 findings against codebase context |
| p1-validator | Validates P1 findings against codebase context |
| p2-validator | Validates P2 findings against codebase context |
Only spawns agents for severity levels that have findings.
| Level | Meaning | Action |
|---|---|---|
| P0 | Critical (security, data corruption, outage) | DO NOT COMMIT - fix immediately |
| P1 | High (logic errors, performance, workflows) | Recommend fixing before commit |
| P2 | Medium (duplications, antipatterns, spec gaps) | Safe to commit, track issues |
| P3 | Low (style, minor improvements) | Optional improvements |
<approval_gates>
Approval Gates
| Gate | Phase | Question |
|---|---|---|
| Scope | 1 | "Confirm files to review?" (if >2000 lines) |
| Remediation | 5 | "Fix P0 now / Fix P0+P1 / Create todos / Report only?" |
</approval_gates>
| Size | Lines | Recommendation |
|---|---|---|
| Ideal | 200-1000 | Fast, thorough |
| Acceptable | 1000-2000 | May take 3-5 min |
| Large | >2000 | Warn user, suggest splitting |
Include: Changed files, related files, test files Exclude: Generated types, node_modules, build artifacts
<critical_rule>
Critical Rule: Report ALL Bugs
NEVER dismiss findings because they are in "pre-existing code".
Report ALL bugs found in reviewed files regardless of when they were introduced. A bug discovered today that was written 3 months ago is still a bug worth fixing.
Invalid reasoning (DO NOT USE):
- "This is pre-existing code unrelated to the current feature"
- "I didn't introduce this bug in this session"
- "This code was written before my changes"
Correct approach:
- Report ALL bugs found in the reviewed files
- Classify by severity (P0-P3) based on impact, not origin
- Let the user decide which to fix </critical_rule>
- Runtime behavior (use manual testing)
- Business logic correctness (requires domain knowledge)
- Test coverage (use test runner)
- Build errors (use typecheck/lint)
- Database migrations (use migration-test-and-push skill)
For comprehensive quality: Run check-your-work + typecheck + lint + tests
<quick_reference>
Quick Reference
Pattern files enforced:
react-typescript-antipatterns.mdCLAUDE.mdzod-form-patterns.mdtanstack-query-patterns.mdperformance-patterns.md</quick_reference>
- references/phase-2-quality-checks.md - All 6 agents for parallel analysis
- references/phase-3-validation.md - Severity validation (up to 3 parallel by level)
- references/consolidated-report.md - Finding interface and report format
<version_history>
Version History
-
v4.2.0 (2026-01-20): Add critical rule for pre-existing bugs
- Report ALL bugs in reviewed files regardless of when introduced
- Fix flawed "not my code" dismissal pattern
-
v4.1.0 (2025-01-18): AI optimization updates
- Add blockquote summary after title
-
v4.0.0 (2025-01-11): Maximum parallelization + unified validation
- Merged Phase 2 + Phase 3 into single Phase 2 (6 parallel agents)
- Phase 3 validation now spawns up to 3 agents (one per P0/P1/P2 level)
- Consistent with check-your-code P0-P3 severity system
-
v3.0.0 (2025-12-28): Refactored to follow skill-authoring-patterns
-
v2.0.0 (2025-10-31): Added severity validation
-
v1.0.0 (2025-10-26): Initial release </version_history>