What it does: Has Codex read your project and write a short AGENTS.md: how to run it, test it, and what not to touch.
Use it when: Codex keeps guessing how your project works and gets it wrong.
You get: An AGENTS.md under 50 lines, plus what Codex couldn't figure out and needs from you.
You are my setup helper with access to my project in Codex, a senior developer who writes the setup notes every new teammate reads first. Goal: read my project and write a short AGENTS.md that says how to run it, test it, and what not to touch, so I have a file Codex reads at the start of each task and it stops guessing. Context: This project runs part of my business. Codex keeps guessing how it works and gets it wrong, because nothing in the project tells it. A short AGENTS.md at the top of the project gives it the same facts every time. First, check that you can read the files in this project. Tell me how many files and top folders you found, and the main language. If you can't read them, stop and tell me to open this project in Codex (steps at simpleai.community/connect/), or ask me to paste the file list instead. Ask me these, then run. Skip any I already answered. 1. What must never change? (Default: .env files and any file with keys or passwords.) 2. Who works on it? (Default: just me, plus Codex.) 3. What is the project, in one line? (Default: guess it from the README and tell me what you guessed.) Do this: 1. Read the project: language, how to install, how to run, how to test, folder layout. For each command, note the file you found it in. 2. If there is already an AGENTS.md, check it first and add to it instead of starting over. 3. Write an AGENTS.md under 50 lines with: - what the project is - the exact commands to install, run and test, only ones you found written in the files - the folder map, one line each - the rules: small changes, run the tests, never touch the files I named, no new libraries without asking 4. List what you couldn't figure out (for example: no test command found) as questions for me. 5. Show me the file. Don't save it until I say yes. Example lines, match this: "Test: npm test (from package.json)" and "Never edit .env. Ask me first." Give me the AGENTS.md in one block under 50 lines, then the open questions as a numbered list. Before you show me, check: - Every command names the file you found it in. If there's no install command written down, write [check: install command]. - The file is under 50 lines, and every file I said must never change is named in it. Rules: - Only commands you can find written in the project's files. - Drafts only. Never commit, push, deploy or delete anything without my yes. - If a script deploys, pushes or deletes, name it in the AGENTS.md rules as never run without my yes. - Use only what is in my data and what I tell you. Never make up a name, number, date, price, link or promise. - If something you need is missing, write [check: what is missing] in that spot and keep going. Last, save these steps as a skill called 'Write AGENTS.md' so you do it this way each time I set up Codex on a project. To run it again, just say: "Run my Write AGENTS.md skill."
You need
CodexTested on sample data
Built on The E-Myth Revisited's rule to build a business that runs on systems, not on you.
Copilot version
There is no separate Copilot version of this one yet. Paste the prompt above into Microsoft 365 Copilot Chat.
See a real example
Tested on sample data Oct 29 2026, after the rewrite: a Claude agent ran it on made-up sample input (tool results simulated). A separate Claude agent judged the run and the card.
Watch: no video for this one yet.
AuthorCorey Leon