Back to Blog

Have Claude Audit Your Claude Setup

claude-code productivity tips

Have Claude Audit Your Claude Setup

Your AI assistant can review its own config — and the suggestions are surprisingly useful

Dotfiles terminal

Definitions (click to expand)
  • ~/.claude — the hidden folder in your home directory where Claude Code stores memories, settings, and global configuration
  • CLAUDE.md — a markdown file Claude reads for persistent instructions about how to behave
  • post-tool-use hook — a command that runs automatically after Claude uses a tool (like editing a file)
  • ESLint — a popular JavaScript/TypeScript linter that catches code style issues and potential bugs

The setup

I’ve been using Claude Code for months, nearly eight hours a day. My CLAUDE.md kept growing — every time I hit a recurring annoyance, I’d add another line. Eventually it was bloated, probably eating more context than it needed to. I wasn’t sure which instructions were actually doing anything.

The trick

Open Claude Code in your ~/.claude directory and ask it to review your setup.

cd ~/.claude && claude

That folder contains your memories, settings, and any global CLAUDE.md you’ve configured. Claude can read all of it, spot patterns you haven’t noticed, and suggest improvements.

In my case, Claude found:

  • My memory wasn’t configured correctly — it wasn’t actually persisting instructions I thought it was saving
  • I kept asking it not to skip tests, which meant I should just put that in CLAUDE.md permanently
  • My config files were getting wordy and needed cleanup
  • I could add a post-tool-use hook to run ESLint automatically instead of fixing lint errors after the fact

What I changed

Claude updated my CLAUDE.md to enforce concise responses:

You are a direct answer assistant.
Rules:
- Sentence length: concise, around 15 words or less
- No fluff: skip greetings, apologies, niceties — start with the answer
- Structure: use bullets/numbers and horizontal lines for visual breaks
- Clarification: if unclear, ask one yes/no question at a time

It also added a post-tool-use hook to run eslint after file edits, which cut down on the back-and-forth lint fixes I was constantly doing.

Bonus: version control your settings

I initialized git inside ~/.claude so I can track every experiment. If a new setting makes Claude behave strangely, I can check the diff or roll back. When you’re tweaking AI behavior daily, having that history is surprisingly handy.

cd ~/.claude && git init

If you’re using Claude Code every day and haven’t pointed it at its own config folder, try it tonight.