Check Your Work
Orchestrate specialized agents to detect bugs, security issues, and quality problems in code changes.
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
| 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
| Gate |
Phase |
Question |
| Scope |
1 |
"Confirm files to review?" (if >2000 lines) |
| Remediation |
5 |
"Fix P0 now / Fix P0+P1 / Create todos / Report only?" |
| 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: 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
- 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
Pattern files enforced:
react-typescript-antipatterns.md
CLAUDE.md
zod-form-patterns.md
tanstack-query-patterns.md
performance-patterns.md
- 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
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
1---2name: check-your-work3description: Orchestrates 6 quality agents in parallel (duplicate-detector, jenny, deep-bug-hunter, security, performance, correctness) with severity validation. Use when user says "check your work", "review what we wrote", "quality check", or after writing new features.4---56# Check Your Work78> Orchestrate specialized agents to detect bugs, security issues, and quality problems in code changes.910<when_to_use>1112## When to Use1314Invoke when user says:1516- "check your work"17- "review what we wrote"18- "quality check the new code"19- "check for bugs in [files/feature]"20- After completing feature implementation21- Before committing significant changes22 </when_to_use>2324<workflow>25## Workflow Overview2627| Phase | Agents | Action |28| ----- | ------------- | -------------------------------------------------------------------- |29| 1 | - | Scope Discovery (identify files) |30| 2 | 6 parallel | All Checks (duplicate-detector, jenny, deep-bug-hunter, 3 reviewers) |31| 3 | up to 3 | Severity Validation (one agent per P0/P1/P2 level with findings) |32| 4 | 1 conditional | Deep Investigation (root cause for validated P0) |33| 5 | - | Consolidated Report + User Decision |3435For Phase 2 details: [references/phase-2-quality-checks.md](references/phase-2-quality-checks.md)36For Phase 3 details: [references/phase-3-validation.md](references/phase-3-validation.md)37For report format: [references/consolidated-report.md](references/consolidated-report.md)38</workflow>3940<agents>41## Agent Summary4243### Phase 2 (6 Parallel)4445| Agent | Focus |46| ----------------------- | --------------------------------------------- |47| duplicate-code-detector | Reimplemented utilities/services |48| jenny | User voice - did we build what was requested? |49| deep-bug-hunter | Logic errors, race conditions, edge cases |50| Security Reviewer | SQL injection, org_id isolation, XSS, PII |51| Performance Reviewer | Memory leaks, bug-causing antipatterns |52| Correctness Reviewer | Logic errors, null handling, async issues |5354### Phase 3 (up to 3 Parallel)5556| Agent | Focus |57| ------------ | ---------------------------------------------- |58| p0-validator | Validates P0 findings against codebase context |59| p1-validator | Validates P1 findings against codebase context |60| p2-validator | Validates P2 findings against codebase context |6162Only spawns agents for severity levels that have findings.6364</agents>6566<severity>67## Severity Classification6869| Level | Meaning | Action |70| ----- | ---------------------------------------------- | ----------------------------------- |71| P0 | Critical (security, data corruption, outage) | **DO NOT COMMIT** - fix immediately |72| P1 | High (logic errors, performance, workflows) | Recommend fixing before commit |73| P2 | Medium (duplications, antipatterns, spec gaps) | Safe to commit, track issues |74| P3 | Low (style, minor improvements) | Optional improvements |7576</severity>7778<approval_gates>7980## Approval Gates8182| Gate | Phase | Question |83| ----------- | ----- | ------------------------------------------------------ |84| Scope | 1 | "Confirm files to review?" (if >2000 lines) |85| Remediation | 5 | "Fix P0 now / Fix P0+P1 / Create todos / Report only?" |8687</approval_gates>8889<scope>90## Scope Guidelines9192| Size | Lines | Recommendation |93| ---------- | --------- | ---------------------------- |94| Ideal | 200-1000 | Fast, thorough |95| Acceptable | 1000-2000 | May take 3-5 min |96| Large | >2000 | Warn user, suggest splitting |9798**Include**: Changed files, related files, test files99**Exclude**: Generated types, node_modules, build artifacts100</scope>101102<critical_rule>103104## Critical Rule: Report ALL Bugs105106**NEVER dismiss findings because they are in "pre-existing code".**107108Report 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.109110**Invalid reasoning (DO NOT USE):**111112- "This is pre-existing code unrelated to the current feature"113- "I didn't introduce this bug in this session"114- "This code was written before my changes"115116**Correct approach:**117118- Report ALL bugs found in the reviewed files119- Classify by severity (P0-P3) based on impact, not origin120- Let the user decide which to fix121 </critical_rule>122123<limitations>124## What This Skill Does NOT Check125126- Runtime behavior (use manual testing)127- Business logic correctness (requires domain knowledge)128- Test coverage (use test runner)129- Build errors (use typecheck/lint)130- Database migrations (use migration-test-and-push skill)131132**For comprehensive quality**: Run check-your-work + typecheck + lint + tests133</limitations>134135<quick_reference>136137## Quick Reference138139**Pattern files enforced**:140141- `react-typescript-antipatterns.md`142- `CLAUDE.md`143- `zod-form-patterns.md`144- `tanstack-query-patterns.md`145- `performance-patterns.md`146 </quick_reference>147148<references>149## References150151- [references/phase-2-quality-checks.md](references/phase-2-quality-checks.md) - All 6 agents for parallel analysis152- [references/phase-3-validation.md](references/phase-3-validation.md) - Severity validation (up to 3 parallel by level)153- [references/consolidated-report.md](references/consolidated-report.md) - Finding interface and report format154 </references>155156<version_history>157158## Version History159160- **v4.2.0** (2026-01-20): Add critical rule for pre-existing bugs161 - Report ALL bugs in reviewed files regardless of when introduced162 - Fix flawed "not my code" dismissal pattern163164- **v4.1.0** (2025-01-18): AI optimization updates165 - Add blockquote summary after title166167- **v4.0.0** (2025-01-11): Maximum parallelization + unified validation168 - Merged Phase 2 + Phase 3 into single Phase 2 (6 parallel agents)169 - Phase 3 validation now spawns up to 3 agents (one per P0/P1/P2 level)170 - Consistent with check-your-code P0-P3 severity system171172- **v3.0.0** (2025-12-28): Refactored to follow skill-authoring-patterns173174- **v2.0.0** (2025-10-31): Added severity validation175176- **v1.0.0** (2025-10-26): Initial release177 </version_history>