Inblog
Start your company blog
GEO

AGENTS.md

AGENTS.md is a Markdown file placed at the root of a repository or website that tells AI agents how to work on the project — build steps, test commands, code style, and rules to respect. Often described as "a README for agents," it has become the de facto standard instruction file, read automatically by OpenAI Codex and other major coding agents at the start of a task.

AGENTS.md is a Markdown file placed at the root of a repository or website that tells AI agents how to work on the project — build steps, test commands, code style, and rules to respect. Often described as "a README for agents," it has become the de facto standard instruction file, read automatically by OpenAI Codex and other major coding agents at the start of a task.

Why It Matters

AI coding agents are powerful but know none of a project's tacit knowledge. Which command runs the tests, which directories are off-limits — all of it had to be restated in every prompt. AGENTS.md pins that context to one file agents reference automatically. It was designed collaboratively by OpenAI Codex, Amp, Jules (Google), Cursor, and Factory, and is supported natively by 20+ tools including GitHub Copilot, Gemini CLI, Devin, and Zed. As of 2026 it is used by more than 60,000 open-source projects, and in December 2025 it was donated — alongside Anthropic's MCP — to the Agentic AI Foundation under the Linux Foundation for neutral governance.

Structure and How to Write One

AGENTS.md is plain Markdown with no required fields. Instead of a fixed schema, it carries natural language instructions an agent can read and follow. Recommended sections:

  • Project overview: A paragraph or two on what the codebase does
  • Build and test commands: Dependency install, dev server, test runner
  • Code style: Naming, formatting, framework conventions
  • Security and boundaries: Files not to touch, how to handle secrets
  • Commit and PR rules: Message format, review process

Large monorepos nest an AGENTS.md per package; agents apply the file closest to where they are working.

How It Differs from llms.txt

If llms.txt is the brochure telling AI what content exists, AGENTS.md is the operating manual explaining how to work here. It started in code repositories, but a newer pattern places it at the website root (/agents.md) to guide browsing agents through a site. Together with llms.txt and Schema.org markup, it falls under the category of "agent-readable signals" — and unlike robots.txt, which governs whether AI crawling is allowed, it covers what an allowed agent should do and how.

GEO Implications

As GEO's center of gravity expands from "content AI cites" to "sites AI can operate," files that give agents explicit instructions are becoming a new optimization layer. If you run developer tools or API docs, the less trial-and-error agents need thanks to AGENTS.md, the more likely your product gets adopted inside agent workflows. Keep in mind the file is guidance, not enforcement — well-structured content remains the foundation.

Sources:

How inblog Helps

If you run a product blog or docs on inblog, remember the division of labor between signals. While llms.txt-style files handle content discovery and AGENTS.md handles working instructions, the blog body itself should be easy for agents to quote and summarize. inblog's clean markup and clear heading structure compound the value of these agent-readable signals.