Your CLAUDE.md Is Making Claude Dumber

You added a CLAUDE.md because you wanted Claude to actually know your project. More rules, more context, more everything. Seemed smart at the time.

Here's the problem. The longer that file gets, the more Claude skims it. And when it skims, it forgets. You didn't build a brain, you built a wall of text nobody reads, including the AI you built it for.

The Fix Is Three Rules, Not Three Hundred Lines

First, cut anything you wouldn't tell a new hire on their first day. If it's not something you'd say out loud to a real person walking into the job, it doesn't belong in the file. Twenty lines that change every answer beat two hundred that Claude ignores.

Second, write instructions, not an encyclopedia. Tone, formats, hard rules, that's what lives in CLAUDE.md. Reference docs, the stuff Claude only needs occasionally, goes in separate files. Not the file it re-reads on every single turn.

Third, one directive per behavior, most important first. This is where most people blow it. They try to be concise AND thorough, like those two things are friends. They're not. Concise and thorough cancel each other out. Pick one. (I know that hurts if you love your bullet points.)

A file Claude skims is worse than no file at all, because you think it's working.

Stop Writing More, Start Writing Less

Go open your CLAUDE.md right now. If it's longer than 20 lines, you've got a skimming problem, not a smart-AI problem. Cut it down to the stuff that actually changes an answer, and move the rest somewhere Claude only visits when it needs to.

Drop a line from yours in the comments on the video and I'll tell you exactly what to cut.

#ClaudeAI #CLAUDEmd #PromptEngineering #AIWorkflows #ClaudeCode