It Tells Claude What Not to Do
A single CLAUDE.md file, built from Andrej Karpathy's notes on where AI coding goes wrong. Drop it into your project and Claude stops over-engineering, stops guessing, and starts double-checking its own work. No dependencies, no setup — just an instruction set.
The Four Rules
Think before coding — state assumptions and ask when the request is ambiguous, instead of silently picking one interpretation. Simplicity first — write the minimum code that solves the task; 50 lines where it would have written 200. Surgical changes — touch only what the task needs and leave working code alone instead of "improving" it. Goal-driven execution — turn the task into a success criterion and loop until it's actually verified.
What's in the Guide
What each rule stops, the plugin commands to add it to every project in seconds, and an important safety note: install it only from the official repo, since a CLAUDE.md from a random cloned repo can carry malicious instructions.
The full breakdown is in the PDF.
Get the Guide
Drop your email below and we'll send it straight to your inbox.