CLAUDE.md generator

Answer a few questions and take away an instruction file that is short, structured and specific. Or paste the one you already have and find out why half of it is being ignored.

Which fileProject nameWhat is it, in a sentence or twoCommands, the thing that cannot be guessedStyle rules that differ from the defaultsOne per line. Leave out anything the language already implies.Rules that are not obvious from the codeThe things you would otherwise explain again next session.NeverHard limits. If one must hold without exception, make it a hook instead.Before you finishThe checklist that closes the loop.

CLAUDE.md

# My project

One or two lines on what this is, for someone who has never seen it.

## Commands

```bash
npm run dev  # development server
npm test     # test suite
```

## Code style

- TypeScript, ES modules, 2-space indentation

## Rules that are not obvious from the code

- Nothing a visitor supplies may be sent to a server

## Before you finish

1. Lint and tests both clean

How it works

The generated file follows the guidance in the official documentation: under 200 lines, grouped under headings, and specific enough that each instruction can be checked.
Commands come first, because they are the one thing an agent genuinely cannot work out by reading your code.
Choose Both and you get an AGENTS.md plus a one-line CLAUDE.md that imports it, since Claude Code reads CLAUDE.md and not AGENTS.md.
The checker reads a file you paste and reports what will cost you: length, vague instructions, directory trees the agent can read itself, and emphasis used so often it stops meaning anything.

Questions

How long should the file be?

Under 200 lines is the documented target. It loads into every session and competes with your actual conversation for attention, so past that the rules start getting lost. The checker tells you where you stand and roughly what it costs per session.

Should I write CLAUDE.md or AGENTS.md?

If your team uses more than one coding tool, put the content in AGENTS.md and have CLAUDE.md import it with @AGENTS.md. If everyone is on Claude Code, a CLAUDE.md alone is simpler. Maintaining both by hand is the option that fails.

Why does the checker complain about my directory tree?

Because the agent can read the filesystem. A tree in the instruction file costs context in every single session and is wrong the first time somebody moves a folder.