An agent rules file

BeginnerRules files

Writes short rules for .cursor/rules, CLAUDE.md, or AGENTS.md, using the repo's real commands.

When to use it

Use it at the start of a repo, or when the agent repeats the same mistake. Do not use it to specify a feature: that is a task prompt, not a permanent rule.

Variables

[PROJECT][STACK][COMMANDS][NEVER][STYLE]

Copy prompt

Write an agent rules file for work in [PROJECT].
Stack: [STACK]
Real commands (test, lint, dev): [COMMANDS]
Never do: [NEVER]
Style: [STYLE]

Format
- I will name the path on the next line if I already know it. If I do not, propose one: AGENTS.md at the root, or CLAUDE.md, or one file under .cursor/rules. Do not create all three.
- 80 lines maximum. Short sentences. No language tutorial.
- Sections: how to run, what you may change, [NEVER], how to know you are done.
- Every command in [COMMANDS] must exist in the repo. If it does not, do not invent the script: write "there is no test command" or ask.
- [STYLE] becomes at most 5 bullets (file names, UI copy language, what not to reformat).
- Do not include a secret, an internal URL with a credential, or "be creative".

Do not
- Do not rewrite the app in this task. Only the rules file.
- Do not delete a block I have marked as a second brain or a human note. If the file already exists, keep that block and show the diff.

Deliver: the full file and a list of what you could not verify in the repo.

How to run this prompt

Point at the file. Say whether you want .cursor/rules, CLAUDE.md, or AGENTS.md. One at a time.
List what never to do. NEVER is the mistake the agent already made in this repo.
Check the commands. If the file cites a script that does not exist, delete the line.

Where this breaks

A long rule file is not read. The prompt caps the length and tells the model to cite the command in package.json, not a generic one.

Result

Slot for a screenshot or demo. No result is published yet.