@zostaff: Andrej Karpathy spent 20 years writing code by hand. Then he switched to Claude Code. A week later he posted everything…

X AI KOLs Timeline News
claude-code andrew-karpathy coding-rules ai-tools viral-repo prompt-engineering

Summary

A set of rules derived from Andrej Karpathy's experience with Claude Code went viral, accumulating 145k GitHub stars in 3 months and offering guidelines to improve AI coding agent behavior.

Andrej Karpathy spent 20 years writing code by hand. Then he switched to Claude Code. A week later he posted everything that broke. Some guy read it and turned it into a 70-line file. 145k stars in 3 months. name: andrej-karpathy-skills 4 rules that fix Claude Code's worst habits: 1. Don't assume. The model picks one interpretation silently and runs with it. This rule forces it to state assumptions and ask when uncertain. 2. Don't hide confusion. If something's unclear, stop. Name what's confusing. Ask. No more "I'll just guess" 3. Surface tradeoffs. Multiple ways to do it? Present them. Don't pick silently. 4. Minimum code that solves the problem. No speculative features. No abstractions for single-use code. If you wrote 200 lines and 50 would do, rewrite it. Drop the file into project root. Works with Claude Code, Cursor, any agent that reads CLAUDE.md. Bookmaked it. Repo below.
Original Article
View Cached Full Text

Cached at: 05/23/26, 12:09 PM

Andrej Karpathy spent 20 years writing code by hand. Then he switched to Claude Code. A week later he posted everything that broke.

Some guy read it and turned it into a 70-line file. 145k stars in 3 months.

name: andrej-karpathy-skills

4 rules that fix Claude Code’s worst habits:

  1. Don’t assume. The model picks one interpretation silently and runs with it. This rule forces it to state assumptions and ask when uncertain.

  2. Don’t hide confusion. If something’s unclear, stop. Name what’s confusing. Ask. No more “I’ll just guess”

  3. Surface tradeoffs. Multiple ways to do it? Present them. Don’t pick silently.

  4. Minimum code that solves the problem. No speculative features. No abstractions for single-use code. If you wrote 200 lines and 50 would do, rewrite it.

Drop the file into project root. Works with Claude Code, Cursor, any agent that reads CLAUDE.md.

Bookmaked it. Repo below.

repo: https://github.com/multica-ai/andrej-karpathy-skills…

trading here: https://predictparity.com/?code=zostaff

Similar Articles

@DivyanshT91162: https://x.com/DivyanshT91162/status/2057692858501804435

X AI KOLs Timeline

Andrej Karpathy's observations on AI coding agent behavior led to the viral CLAUDE.md file, which provides 4 behavioral rules for AI agents and became one of the fastest-growing repositories on GitHub, signaling a shift from AI intelligence to AI discipline in coding.

@yaohui12138: Karpathy released a GitHub open-source project that truly amazed me. The project is called andrej-karpathy-skills, with 130k+ stars on GitHub. I'd call it the most useful AI engineering project of 2026. The problem it solves is extremely precise: making Cl…

X AI KOLs Timeline

Karpathy released an open-source project called andrej-karpathy-skills, centered around a 4KB CLAUDE.md file containing 4 behavioral guidelines (Think Before Coding, Simplicity First, Surgical Changes, Goal-Driven Execution). It significantly reduces AI coding error rates (up to 90%), improving code quality and development efficiency.