What's this about?
The author of the Game Engine Black Book series published the agent instruction file he uses to keep LLM-assisted code quality up — a concrete, opinionated example rather than a general essay about prompting.
What exactly is an agent.md file?
It is a plain text file in your repository containing standing instructions for any AI agent working in that codebase: conventions to follow, things never to do, how to run the tests. The agent reads it before touching anything. It is the closest thing the field has to a coding standards document that the machine actually obeys.
What's the community saying?
HN 413 points, 175 comments
- Strong appetite for concrete examples over advice — the value here is that it is a real file from a real practitioner.
- Debate over whether rules like these actually hold across a long session or get gradually forgotten as context fills.
- It ties into the Claude Code hackability item above, since AGENTS.md support is precisely what is being requested of vendors.