A developer named Forrest Chang was reading notes from Andrej Karpathy — one of the people who helped build the original ChatGPT — about the annoying ways AI coding assistants tend to misbehave. So he turned those observations into a single small file you drop into your project, and suddenly Claude Code starts acting differently.
The file corrects four specific bad habits: making silent assumptions instead of asking, over-engineering simple things, quietly deleting code it doesn't understand, and never telling you when there are real trade-offs to consider.
The result? Nearly 6,000 people starred it on GitHub within a single day. That kind of reaction is rare — it means a lot of people recognized the problem immediately.
If you're working with a developer who uses Claude Code, or if you've started experimenting with it yourself, this file is the difference between an AI assistant that confidently goes off in the wrong direction and one that pauses, flags uncertainty, and keeps things simple.
You don't need to understand how it works. You just need to know it exists — and mention it to whoever is building things for you.
Find it here: https://github.com/forrestchang/andrej-karpathy-skills
Claude Code — Anthropic's AI assistant designed specifically to help write and edit software. Think of it as an intern who can type very fast.
CLAUDE.md — A plain text file that gives Claude a set of rules to follow whenever it works on a particular project. Like leaving a note for a contractor before they start work.
GitHub stars — A rough measure of how much the developer community cares about something. 6,000 in a day is a standing ovation.