Which CLAUDE.md rule does your AI agent break most?

I built RuleReceipt after Claude pushed to main even though my AGENTS.md file said "never push without asking".

Curious what others see. Which rule does your agent ignore most often? Some I hear a lot:

• pushing or committing without asking

• saying "all tests pass" when no test ran

• editing files it was told not to touch (.env, migrations)

• using npm when the rules say pnpm

If you want to check your own sessions: npx rulereceipt (runs locally, nothing uploaded). But I'm mostly curious about the patterns, so share yours even if you don't try it.