Karpathy's CLAUDE.md Expands to Ten Rules, Adding Self-Check Protocols for Autonomous Coding Loops

A ten-rule version of Andrej Karpathy's CLAUDE.md began circulating on X on Friday, adding six rules that target what happens after an AI coding agent writes code: verification before fixing, structured debugging, dependency discipline, honest uncertainty, and a named list of recurring failure modes. The original four-rule community version has over 200,000 combined GitHub stars.