A collection of Claude Code skills for comprehensive code auditing. These skills work together as a layered system: orchestrators compose individual audits, two-pass sweeps generate tests then fix issues, and component skills provide focused analysis.
Copy the skills/ directory into your Claude Code skills folder:
cp -r skills/* ~/.claude/skills/Or symlink individual skills:
ln -s "$(pwd)/skills/sweep1" ~/.claude/skills/sweep1 Orchestrators
(compose audits in parallel)
┌──────────────┬──────────────┬──────────────┐
│ /sweep1 │ /sweep2 │ /sweep3 │
│ Contracts │ Quality │ Ops & │
│ & Security │ │ Infra │
└──────┬───────┴──────┬───────┴──────┬───────┘
│ │ │
┌──────┴──────┐┌──────┴──────┐┌──────┴──────┐
│/api-contract││ /tech-debt ││/performance- │
│ -audit ││ ││ profile │
│/migration- ││/style-review││/dependency- │
│ review ││ ││ audit │
│/security- ││/test-health-││/production- │
│ audit ││ audit ││ ready │
└─────────────┘└─────────────┘└─────────────┘
Component Skills (read-only analysis)
┌─────────────────────────────────────────────┐
│ Two-Pass Sweeps │
│ Pass 1: audit agent (read-only, generates │
│ tests + .tmp analysis file) │
│ Pass 2: fix agent (runs tests, fixes code, │
│ 3-attempt loop, writes docs) │
├─────────────┬───────────────┬───────────────┤
│/architecture│ /crud-sweep │ /fuzz-sweep │
│ -sweep │ │ │
│/state- │ /category- │ │
│ machine- │ sweep │ │
│ sweep │ │ │
└─────────────┴───────────────┴───────────────┘
┌─────────────────────────────────────────────┐
│ Foundational │
│ │
│ /design-patterns — GOF vocabulary │
│ (used by all sweeps) │
│ /add-events — event-sink refactoring │
│ (prerequisite for /fuzz)│
│ /production-sinks — file sinks for events │
│ (logs/catastrophic.log, │
│ logs/errors.log) │
└─────────────────────────────────────────────┘
These run /design-patterns first to establish vocabulary, then launch three audits in parallel, then run /design-patterns again to synthesise findings.
| Skill | Focus | Component Audits |
|---|---|---|
/sweep1 |
Contracts & Security | /api-contract-audit + /migration-review + /security-audit |
/sweep2 |
Code Quality | /tech-debt + /style-review + /test-health-audit |
/sweep3 |
Ops & Infrastructure | /performance-profile + /dependency-audit + /production-ready |
These launch two agents sequentially: a read-only audit agent that generates tests and writes a .tmp analysis file, then a fix agent that runs the tests, fixes code, and retries up to 3 times.
| Skill | What it does | Output |
|---|---|---|
/architecture-sweep |
Module depth, information hiding, dependencies, event-sink separation (Ousterhout) | ARCHITECTURE.md |
/crud-sweep |
Maps all CRUD operations against a 40+ edge case matrix, generates integration tests | CRUD coverage report |
/fuzz-sweep |
Discovers system boundaries, generates fast-check fuzz tests. Prerequisite: /add-events |
Fuzz test report |
/state-machine-sweep |
Detects status fields as finite automata, generates property-based random walk tests | State machine report |
/category-sweep |
Finds functors, monads, Kleisli categories, Yoneda instances; writes law tests | CATEGORY_THEORY.md |
These analyse and report but never modify code.
| Skill | Focus |
|---|---|
/design-patterns |
GOF pattern inventory: patterns in use, anti-patterns, recommended introductions |
/api-contract-audit |
Response consistency, backward compatibility, idempotency, schema evolution, error contracts |
/migration-review |
Destructive ops, locking risks, missing indices, rollback strategy |
/security-audit |
OWASP Top 10, injection, auth bypass, input validation, data exposure |
/tech-debt |
Coupling, god objects, dead code, duplicated logic; prioritised debt register |
/style-review |
Clarity, obviousness, module depth, naming, exception handling |
/test-health-audit |
Pyramid shape, coverage gaps, test quality, flakiness, maintainability |
/performance-profile |
N+1 queries, bundle bloat, memory leaks, missing caches |
/dependency-audit |
CVEs, outdated packages, unused deps, license conflicts, supply chain risks |
/production-ready |
Observability, reliability, data integrity, performance, concurrency, deployment safety |
| Skill | Purpose |
|---|---|
/add-events |
Refactors business logic to return events as pure data (event-sink pattern). Prerequisite for /fuzz-sweep. |
/production-sinks |
Creates file sinks that route events by severity to logs/catastrophic.log and logs/errors.log (JSONL). Companion to /add-events. |
Run any skill from Claude Code:
/sweep1 # Full contracts & security audit
/architecture-sweep # Two-pass architecture review
/crud-sweep # CRUD edge case audit
/fuzz-sweep api # Fuzz API routes only
/state-machine-sweep discover # Find all state machines
/category-sweep full # Full categorical analysis
/design-patterns src/lib # GOF analysis of a specific module
/tech-debt full # Full tech debt assessment
/dependency-audit security # Security vulns only
Before shipping:
/sweep1 # contracts, migrations, security
Before refactoring:
/sweep2 # tech debt, style, test health
Before deployment:
/sweep3 # performance, deps, production readiness
Full audit (all three):
/sweep1
/sweep2
/sweep3
Event-driven testing pipeline:
/add-events # 1. Refactor to return events
/production-sinks # 2. Create file sinks for errors
/fuzz-sweep # 3. Fuzz all boundaries
/state-machine-sweep # 4. Test state machines
/crud-sweep # 5. Test CRUD edge cases
-
Pass 1 (Audit Agent): Read-only. Analyses the codebase, generates test files, writes a
.tmpanalysis file (e.g.,architect.tmp,crud-audit.tmp). Never modifies application code. -
Between Passes: The orchestrator confirms the
.tmpfile exists and summarises findings. -
Pass 2 (Fix Agent): Reads the
.tmpfile, runs the generated tests, diagnoses failures, fixes application code, retries up to 3 times. Deletes the.tmpfile when done. May write documentation (e.g.,ARCHITECTURE.md).
The /security-audit skill is currently scoped to a healthcare CRM (Next.js/Prisma). To adapt it for your project, update the "Critical Surfaces" section in skills/security-audit/SKILL.md to list your project's attack surfaces.
All other skills are project-agnostic and auto-detect the tech stack.
MIT