audit-skill
adewale/audit-skill
Comprehensive audit toolkit with 14 audit types for Claude Code — branch audits with Clean/Minor/Blocking verdicts plus deep-dive audits for concurrency, security, performance, and more
Instalação
npx convez-ai adewale/audit-skill1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
README
A tradução para o português deste README ainda está na fila.
Audit Skill
Audit Skill is an Agent Skill with 14 audit types. Catches issues before you push, and goes deep when you need it.
What it does
Branch audit (default) — Reviews your branch diff against 8 categories and gives a Clean/Minor/Blocking verdict:
- Secrets and credentials
- Unintended changes
- Debug artifacts
- Test coverage
- Build and suite
- Commit hygiene
- Integration check
- Merge conflicts and rebase state
Deep-dive audits — 13 project-wide analyses using sub-agents:
- Code quality — duplication, inconsistency, simplification opportunities
- Documentation brittleness — fragile references, over-specified details, staleness risk
- Documentation-code sync — docs that have already drifted from the code
- Language best practices — idiomatic patterns for Python, JS/TS, Go, Rust, Java, Ruby, Shell
- Concurrency — shared mutable state, race conditions, deadlock risk, thread/goroutine leaks
- Resource management — file handle leaks, unclosed connections, zombie processes, temp files
- Test quality — weak assertions, flaky patterns, test isolation, missing negative tests
- Feature completeness — documented features vs actual implementation, missing endpoints
- Performance — N+1 queries, unbounded caches, hot-path allocations, blocking I/O
- Bug patterns — shallow merges, serialization mismatches, silent data loss, stale closures
- Design philosophy compliance — evaluate code against the project's own stated principles
- Security vulnerabilities — injection, auth, data exposure, dependency risks, config
- UI design — review using CRAP principles (Contrast, Repetition, Alignment, Proximity)
Installation
Via Claude Code plugin marketplace
/plugin marketplace add adewale/audit-skill
/plugin install audit@adewale-audit
Via skills.sh
npx skills add adewale/audit-skill
Skills appear on skills.sh automatically after users install the repo with the skills CLI. Install counts and leaderboard rankings come from anonymous CLI telemetry; opt out with DISABLE_TELEMETRY=1. The repo page customization in skills.sh.json is picked up after the repository is seen by telemetry and the cache refreshes.
Agent compatibility
The installable skill directory is skills/audit. It uses the Agent Skills SKILL.md format and is configured for Codex, OpenCode, Pi, Gemini CLI, and Claude Code.
| Agent/client | Install or use |
|---|---|
| Codex | cp -R skills/audit ~/.codex/skills/audit |
| OpenCode | cp -R skills/audit ~/.config/opencode/skills/audit or use .opencode/skills/audit in a project |
| Pi | pi install https://github.com/adewale/audit-skill or pi --skill skills/audit |
| Gemini CLI | gemini skills install https://github.com/adewale/audit-skill --path skills/audit or copy to .gemini/skills/audit |
| Claude Code | npx skills add adewale/audit-skill or copy to .claude/skills/audit |
Manual
Copy the canonical skill directory into your agent's skills directory:
gh repo clone adewale/audit-skill
cp -R audit-skill/skills/audit ~/.claude/skills/audit
cp -R audit-skill/skills/audit ~/.codex/skills/audit
cp -R audit-skill/skills/audit ~/.config/opencode/skills/audit
Usage
Invoke with /audit where slash commands are available, or ask your agent directly:
> /audit
> audit
> review my changes before I push
> check what I'm about to push
> run a concurrency audit
> audit test quality
> do a security audit
> check for performance issues
> are our docs up to date with the code?
> review the UI design
Example output
Branch audit — Blocking verdict
## Summary
| # | Category | Finding | Severity |
|----|-------------------------|------------------------------------------------|----------|
| 1 | Secrets and credentials | Live Stripe secret key hard-coded in source | Blocking |
| 2 | Debug artifacts | print() debug statement on line 7 | Blocking |
| 3 | Test coverage | No tests for any new payment code | Blocking |
Verdict: **Blocking** — issues that should be fixed before pushing.
Branch audit — Clean verdict
Verdict: **Clean** — no findings, safe to push.
Eval results
Tested across 6 scenarios with and without the skill:
Branch audits (3 scenarios: dirty/clean/mixed branches):
| Configuration | Pass rate | Avg time | Avg tokens |
|---|---|---|---|
| With skill | 93% | 112s | 21,619 |
| Without skill | 73% | 94s | 17,717 |
Deep-dive audits (3 scenarios: concurrency/test quality/performance):
| Configuration | Pass rate | Avg time | Avg tokens |
|---|---|---|---|
| With skill | 100% | 131s | 31,383 |
| Without skill | 100% | 133s | 24,219 |
The skill's primary value is consistent structured output — Clean/Minor/Blocking verdicts, systematic checklists, and organized report formats. Detection of issues is similar either way since Claude is already good at finding problems. The skill ensures they're reported in a reliable, repeatable format.
License
MIT