
Karpathy Skills
One CLAUDE.md distilled from Andrej Karpathy's notes on LLM coding pitfalls — think first, keep it simple.
About
Karpathy Skills (by multica-ai) is a single CLAUDE.md file designed to improve Claude Code behavior, derived from Andrej Karpathy's widely-shared observations on LLM coding pitfalls. It encodes four principles that directly counter the failure modes Karpathy described: Think Before Coding (wrong assumptions, hidden confusion, missing tradeoffs), Simplicity First (overcomplicated code, bloated abstractions, 1000 lines where 100 would do), Surgical Changes (not touching code or comments it doesn't understand as a side effect), and Goal-Driven Execution (tests-first, verifiable success criteria). It is one of the most-starred single-file skill repos (~200k stars), installable via skills.sh or as a Claude Code plugin, with English and Chinese READMEs.
The honest picture: it is literally one file of behavioral guidelines — powerful as a corrective lens, but no substitute for domain-specific skills, tests, or review. Its effectiveness depends on the agent actually reading and honoring CLAUDE.md, which varies by harness. Because it is so simple, unofficial forks and copies abound; install from the genuine multica-ai/andrej-karpathy-skills source.
What it can do
- Think-before-coding discipline
- Simplicity-first code guidance
- Surgical, minimal-diff changes
- Tests-first, goal-driven execution
Setup
Add the CLAUDE.md to your project root or install via skills.sh / Claude Code plugin marketplace.
Prerequisites
Claude Code or another agent that reads CLAUDE.md files.
Use cases
Works with
Trust score — 9/10
Scored 9/10 (verified tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 50; Trending score 88/100 from researcher assessment.
★ 204,000 GitHub stars
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.