A collection of specialist Claude Code agents — each with a distinct persona, domain expertise, and focus. Some review code for risks, others explain codebases, and others synthesize feedback. Together they provide thorough, multi-perspective coverage across the development lifecycle.
| Agent | File | Mission | Prevents |
|---|---|---|---|
| 🔎 Fact Checker | fact-checker.md |
Verify claims against reality | "Sounds right" engineering |
| 🔢 Numbers Man | numbers-man.md |
Numerical correctness everywhere | Silent math bugs |
| ⚖️ Compliance Counsel | compliance-counsel.md |
Legal, privacy, accessibility | Long-tail legal risk |
| 🔍 5 Whys Analyst | five-whys-analyst.md |
Root cause identification | Band-aid engineering |
| 🌐 Systems Thinker | systems-thinker.md |
Cross-service ripple effects | Local fix, global instability |
| 😈 Devil's Advocate | devils-advocate.md |
Challenge consensus and rationale | Groupthink |
| 🔓 Ethical Hacker | ethical-hacker.md |
Security vulnerability discovery | Production security incidents |
| 💥 Chaos Engineer | chaos-engineer.md |
Failure and resilience testing | Fragility under load |
| ♿ Usability Champion | usability-champion.md |
Inclusive, accessible design | Unusable software |
| 👶 First-Time User | first-time-user.md |
New user perspective | Insider bias |
| 🎪 Mr Playful | mr-playful.md |
Creative edge case discovery | Obscure edge cases |
| 📊 Value Guardian | value-guardian.md |
Business alignment and ROI | Shipping busywork |
| 📈 Scalability Auditor | scalability-auditor.md |
Growth readiness | Predictable rewrites |
| 📝 Documentarian | documentarian.md |
Documentation completeness | Knowledge silos |
| 🧪 Test Skeptic | test-skeptic.md |
Test quality and confidence | False confidence |
| 🚀 Migration Guardian | migration-guardian.md |
Deployment and rollback safety | Deployment nightmares |
| ⚡ Performance Profiler | performance-profiler.md |
Runtime performance | Death by 1000 cuts |
| 🗺️ Code Sherpa | code-sherpa.md |
Codebase explanation | Cargo-cult coding |
| 🧩 Moderator | moderator.md |
Cross-agent synthesis | Review paralysis |
| 📋 Performance Reviewer | performance-reviewer.md |
Review feedback synthesis | Vague, evidence-free reviews |
Best for: Ensuring correctness, accuracy, and compliance.
- Fact Checker — Is what we claim true?
- Numbers Man — Is the math right?
- Compliance Counsel — Is it legal and accessible?
Best for: Post-incident review, architectural changes, bug fixes.
- 5 Whys Analyst — What's the real cause?
- Systems Thinker — What else does this affect?
Best for: Security-sensitive features, critical paths, infrastructure changes.
- Devil's Advocate — Should we do this differently?
- Ethical Hacker — Can this be exploited?
- Chaos Engineer — What happens when things fail?
Best for: UI changes, new features, onboarding flows.
- Usability Champion — Can everyone use this?
- First-Time User — Can a newcomer understand this?
- Mr Playful — What weird thing will a real user do?
Best for: Feature planning, architecture decisions, technical investment.
- Value Guardian — Does this matter?
- Scalability Auditor — Will this survive growth?
Best for: PRs, deployment preparation, test review.
- Documentarian — Can the next person understand this?
- Test Skeptic — Do the tests actually prove anything?
- Migration Guardian — Can we deploy and rollback safely?
- Performance Profiler — Is it fast enough right now?
Best for: Joining a new project, onboarding, navigating unfamiliar code.
- Code Sherpa — What does this code actually do and why?
Best for: Running after multiple agents, large reviews, conflicting feedback.
- Moderator — What do we actually do with all this feedback?
Best for: Performance review cycles, feedback synthesis.
- Performance Reviewer — Turn raw feedback into evidence-based review responses
Different situations call for different agent combinations:
Run: Fact Checker + Test Skeptic + Performance Profiler Add if UI changes: Usability Champion + First-Time User Add if API changes: Migration Guardian + Systems Thinker
Run: Ethical Hacker + Compliance Counsel + Chaos Engineer Follow up with: Moderator (to synthesize and prioritize)
Run: Value Guardian + First-Time User + Mr Playful + Scalability Auditor Follow up with: Documentarian + Migration Guardian
Run: 5 Whys Analyst + Systems Thinker + Chaos Engineer Follow up with: Test Skeptic (to verify the test suite would now catch this)
Run: Devil's Advocate + Systems Thinker + Scalability Auditor + Value Guardian
Run: Migration Guardian + Numbers Man + Chaos Engineer
Run: Code Sherpa (full codebase overview) Follow up with: Code Sherpa (targeted walkthroughs of key features) Add before first PR: Documentarian (to flag gaps the new developer noticed)
See the Agent Template section in CLAUDE.md for the full template. The key ingredients:
- Strong identity — "You are an expert..."
- Clear mission — one overriding constraint
- Specific focus areas — actionable, not vague
- Typical questions — primes investigative instincts
- Methods — step-by-step operational workflow
- Guardrails — what NOT to do (prevents overreach)
- Failure mode — the category of failure this agent prevents
- Scope — when to activate, what to engage with
- Output format — consistent, actionable output (reports for reviewers, explanations for explainers, etc.)
Place agent .md files in your Claude Code project's .claude/agents/ directory (or wherever your Claude Code configuration expects custom agents). Each agent can then be invoked by name during Claude Code sessions.