Agent skills & prompts
multica-ai/andrej-karpathy-skills
A single CLAUDE.md with four principles to improve Claude Code
- 203k
- No OSI license
What it is and why it matters
It is a single CLAUDE.md file that codifies Andrej Karpathy's observations on typical LLM coding pitfalls. It boils down to four principles: think and ask for clarification before coding, prioritize simplicity, make surgical changes, and run goal-driven. Its appeal is how simple it is: you drop it into your setup, and it aims to reduce wrong assumptions and over-engineering.
Key points
- Pushes the agent to ask and clarify intent before it starts building
- Encourages the minimum amount of code and surgical changes that leave untouched what you did not ask for
- Installs as a single file; a detailed prompt is advisable to integrate it if you already have a CLAUDE.md
Where we saw it
Jay E | RoboNuggets 2026-04-16 Compares two Claude Code sessions, with and without the file, on the same app: the Karpathy-principles version actually implements the light-theme toggle and the search filter with fewer lines of code, while the vanilla version claims to have done it but does not reflect it. The Karpathy CLAUDE.md File That 43,000 Developers Installed in 1 Week (Full Breakdown) The Next New Thing 2026-04-17 The hosts highlight it as a repo that remains very popular and easy to use: someone turned Karpathy's tweet into a repository with the principles codified, worth including if you are building. 🔥Hot Repos Explained
Same category