👋 Hi Product Hunt! I build RuleReceipt: proof of whether your AI coding agent followed your rules

Most intros start with a bio. Mine starts with a number: 63%.

I parsed 559 public CLAUDE.md files (the rules people write for their AI coding agents). 63% of what's in them isn't an instruction at all: folder trees, notes, docs. And of the real rules, under half are concrete enough to ever check automatically.

That explained something that had been bugging me. My rules file said "never push without asking", and Claude pushed to main anyway. Nothing told me. The rules were advice, and nobody was checking.

So I built RuleReceipt. It reads your real agent sessions and tells you, rule by rule: followed, broken, or can't tell, with the exact line as proof. It only says "broken" with hard evidence. Runs locally; nothing leaves your machine.

What I've learned building it: a checker that cries wolf is worse than no checker. Most of the work went into not accusing wrongly, not into catching more.

We're launching today, and it's Hypership Day, so if you try it and something's missing, tell me and I'll ship the small fixes today. 🚢

Happy to talk about AI agents, rules files, or building dev tools solo. What's your experience: does your coding agent actually follow your rules?