AGENTS.md: a README for the robots

AI coding agents are becoming regular contributors to our codebases — and like any contributor, they work better with good onboarding. The emerging answer is AGENTS.md: an open convention for a dedicated instruction manual for agents, separate from the README.

The vendor-specific alternatives — a copilot-instructions.md here, a CLAUDE.md there — fragment the same information across files that each tool ignores except its own. A single, predictable entry point is better for everyone.

Why it works

• README.md stays clean and focused on people. Humans get the narrative: what the project is, why it exists, how to get started.
• Agents get a predictable entry point. Build commands, conventions, constraints and gotchas, in one place every tool can find.
• Machine-readable guidance stops cluttering the main docs. The instructions an agent needs ("always run lint before committing", "never touch this directory") are noise to a human reader.

---

This site practices what this note preaches: there's an AGENTS.md at the repo root, and CLAUDE.md is just a symlink to it. The robots and the humans each get their manual.