convez.ai
Skill
adewale

audit-skill

adewale/audit-skill

Ver no GitHub
InicialPluginScore · de 0 a 100

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-skill

1 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

Estrelas
3
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 1 semana
Versão
v1.0.2
Issues abertas
4
Contribuidores
1

README

A tradução para o português deste README ainda está na fila.

Audit Skill

skills.sh

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/clientInstall or use
Codexcp -R skills/audit ~/.codex/skills/audit
OpenCodecp -R skills/audit ~/.config/opencode/skills/audit or use .opencode/skills/audit in a project
Pipi install https://github.com/adewale/audit-skill or pi --skill skills/audit
Gemini CLIgemini skills install https://github.com/adewale/audit-skill --path skills/audit or copy to .gemini/skills/audit
Claude Codenpx 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):

ConfigurationPass rateAvg timeAvg tokens
With skill93%112s21,619
Without skill73%94s17,717

Deep-dive audits (3 scenarios: concurrency/test quality/performance):

ConfigurationPass rateAvg timeAvg tokens
With skill100%131s31,383
Without skill100%133s24,219

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

Ver o original no GitHub

Bom para

Relacionados