andrej-karpathy-skills
GitHub Repo Pretty sure · Works if Claude reads itA CLAUDE.md file that packages common sense about LLM coding into four principles. Addresses real problems Karpathy identified. Utility depends entirely on whether your AI actually reads it.
Agent rating
Agent reasoning
This is a well-reasoned prompt engineering artifact, not a technical contribution. The four principles (Think Before Coding, Simplicity First, Surgical Changes, Goal-Driven Execution) directly address documented LLM failure modes — wrong assumptions, overengineering, orthogonal edits. The slop score reflects that it's a markdown file asking AI to behave better, not novel methodology. Signal is high because the advice is specific and actionable, and it genuinely improves Claude Code output in ...
Become a MFer to rate — log in