Every AI setup drifts. You add a skill, rename another, update one context file and forget the two that mention it. After a few weeks, your instructions quietly disagree with each other and the assistant has to guess which one wins.
I hit this with my own system. The skill that drew my system map still pointed at agent files I had deleted. So I rebuilt it as an audit that works out what exists instead of trusting a list.
What it does
It scans your Claude setup: CLAUDE.md and AGENTS.md files, skills, agents, rules, commands, hooks, MCP servers, and the context folders your instructions point to. Then it gives you two things:
- A one-page map (
ai-system-map.html) showing how everything connects, with a card for each item. - A short "fix first" list, ranked by impact.
What it checks
- Broken: references to files, skills or scripts that no longer exist.
- Stale: "last updated" dates that are months old, or indexes listing things that are gone.
- Conflicting: two files giving different instructions for the same thing, quoted side by side.
- Overlapping: skills with descriptions so similar the assistant could pick either.
- Possibly unused: files nothing points to.
- Cost and safety: always-loaded instructions that have grown long, and anything that looks like a secret. It reports the file and line, never the value.
It only reads. Nothing changes unless you pick a finding and ask it to fix it.
Try it
- "Audit my AI system"
- "Map my Claude setup, project only"
- "What's stale in my skills?"
Run it after adding or renaming skills, or once a month.
Install
In Claude Code, copy the folder into your skills directory:
git clone https://github.com/ecemkaraman/ai-skills.git
cp -r ai-skills/skills/ai-system-audit ~/.claude/skills/
In the Claude app, zip the ai-system-audit folder, then go to Customize → Skills, click +, choose Create skill → Upload a skill, and upload the zip. Code execution needs to be on in Settings → Capabilities.