Building RuleReceipt — a CLI that checks whether an AI coding agent actually followed the rules in your CLAUDE.md / AGENTS.md, with quoted evidence. Runs locally, zero network calls by default.Ran it against a real session of my own and it reported 10 rule violations. All 10 were false. Write-up of what was structurally wrong: https://rulereceipt.dev/postmortem
founder@rulereceipt.dev