The 100k‑Star CLAUDE.md Rule File That Eliminates Common AI Coding Bad Habits

Problem

AI coding agents such as Claude or Cursor often make unchecked assumptions, add unnecessary abstractions, and modify code they do not fully understand, leading to messy diffs and rework. Andrej Karpathy publicly noted these behaviors.

Solution – CLAUDE.md

A community‑maintained CLAUDE.md file (GitHub https://github.com/forrestchang/andrej-karpathy-skills, >100 k Stars) encodes a concise set of guidelines that steer AI assistants toward disciplined engineering practices.

Four Core Principles

1. Think Before Coding

Surface ambiguities, list possible interpretations, and ask for clarification before proceeding. Push back when a simpler solution exists. Explicitly state uncertainty.

2. Simplicity First

Write the smallest code that solves the current problem. Do not add unused functionality or abstractions. Keep implementations short (e.g., 200 lines instead of 1 000).

3. Surgical Changes

Modify only the code directly related to the task. Preserve existing style, avoid touching unrelated files, and remove only imports introduced by the change.

4. Goal‑Driven Execution

Turn tasks into verifiable goals. Write failing test cases for invalid inputs or bugs, then iterate until all tests pass.

Users report cleaner initial code, dramatically fewer rework cycles, and smaller diffs.

Installation

For Claude Code users, add the plugin marketplace and install the plugin:

# Add plugin marketplace
/plugin marketplace add forrestchang/andrej-karpathy-skills
# Install plugin
/plugin install andrej-karpathy-skills@karpathy-skills

Per‑project installation (run in project root):

curl -o CLAUDE.md https://raw.githubusercontent.com/forrestchang/andrej-karpathy-skills/main/CLAUDE.md

After installation the rules apply automatically to all projects.

Reference

Repository: https://github.com/forrestchang/andrej-karpathy-skills