@anyelamarillo - Karpathy's 65-Line CLAUDE.md Rules
@anyelamarillo on X Andrej Karpathy's CLAUDE.md file, just 65 lines with four rules for guiding AI coding assistants, hit #1 on GitHub Trending with over 220,000 stars. The post claims it raised AI coding accuracy from 65% to 94%.
- The four rules: think before coding (state assumptions, ask when unsure), simplicity first (minimal code, no unrequested abstractions), surgical changes (touch only what the request covers), and goal-driven execution (turn vague instructions into verifiable success criteria).
- The post attributes a jump from 65% to 94% accuracy to the file, though it gives no benchmark or methodology behind that figure.
Karpathy's CLAUDE.md, a 65-line instruction file for AI coding assistants, reportedly hit #1 on GitHub trending with over 220,000 stars. The post, written in Spanish by Anyela Marillo, claims it lifted AI coding accuracy from 65% to 94%. Most of that is a social media pitch, and the 65-to-94 figure comes with no benchmark, test setup, or source attached, so treat it as the author's claim rather than a measured result. The file itself is the more useful part.
It rests on four rules. First, think before coding: state assumptions, ask when unsure, and don't guess. Second, favor simplicity: write the least code that solves the problem and skip abstractions nobody requested. Third, make surgical changes: don't touch unrelated code, and every changed line should trace back to the request. Fourth, goal-driven execution: turn vague instructions into verifiable success criteria before writing anything. These are sensible habits for working with a model that tends to over-engineer, invent requirements, and rewrite more than it was asked to. Anyone who has watched an assistant refactor half a file to fix one bug will recognize the problem.
The practical takeaway is that the rules are cheap to try and easy to adapt. A 65-line instruction file costs nothing to test on your own codebase, and you can compare diff size and rework before trusting the headline number. Save it if you use AI coding tools, but verify the gains yourself.