TL;DR
- Check out https://github.com/openai/agents.md
- Migrate your existing AI agent instruction files
- create an AGENTS.md file at the root of your repository
- other AI instruction files are symbolic links to the AGENTS.md at the root level of the project
Where does this proposal come from?
Originally, the Amp Code team started supporting an AGENT.md file for their AI coding agents (https://ampcode.com/news/AGENTS.md). Shortly later, OpenAI picked that up but in the plural form. As stated on the agents.md project website, AGENTS.md “emerged from a collaborative effort across the AI software development ecosystem”.
Why?
Instead of maintaining separate configurations, you only maintain one. This was the case already if you used the mechanism outlined in my previous article “AI Instructions in Code Repositories”. But now the file at root level has the naming convention “AGENTS.md”, which allows companies following this open format to offer built-in support for that file in their agent.
It saves you the time of telling your coding agent which file to use.