claudeaudit

A skill that audits your repository for AI agent readiness. How well can Claude Code understand, run, verify, and scale across your codebase?

Fourteen categories across two dimensions: config readiness (scored with a maturity level) and ergonomics (how productive agents are in the codebase). Works as an installable skill or just tell your agent to fetch this page.

Install

$ npx skills add heimann/claudeaudit

Or just tell Claude to run it

fetch claudeaudit.dev/SKILL.md and run it against this repo

Supports Claude Code, Codex, Cursor, and more.

Example output

claudeaudit - acme/billing-service Ok (permissions blocks Safe) docs 2 structure 3 bootstrap 3 tests 2 perms 1 hooks 2 rules 2 worktree 1 sandbox 2 feedback 2 errors 1 types 2 determinism 2 locality 2 Top fixes: 1. Add .claude/settings.json with allowedTools for mix test/format/credo -> perms 1->2, unlocks Safe 2. Make PORT configurable via env var in config/dev.exs -> worktree 1->2, toward Great 3. Configure --formatter for concise test output -> errors 1->2

What it checks

Readable

Agent documentation - CLAUDE.md quality, sub-directory docs

Repository structure - layout clarity, entry points, boundaries

Runnable

Dependency bootstrapping - clone to running in one command

Self-verification - tests, E2E coverage, seed data

Safe

Permissions - .claude/settings.json, scoped tool access

Code quality hooks - pre-commit, lint, format automation

Rules & conventions - patterns, naming, architecture decisions

Scalable

Worktree readiness - no hardcoded paths, configurable ports

Sandbox compatibility - containerized dev, mockable deps

Ergonomics

Feedback loop - scoped tests, watch mode, fast verification

Error signal quality - concise errors, not noisy stack traces

Type system - static analysis catches mistakes before runtime

Determinism - same command, same result, no flaky tests

Change locality - change one thing without reading everything