AGENTS.md is a simple, open-source file format designed to guide AI coding agents by providing essential context, instructions, and project-specific conventions. By acting as a 'README for agents,' this format enables developers to share build steps, testing protocols, code styles, and other technical details that agents need to operate effectively, keeping the primary README.md focused on human contributors. The project emerged from a collaborative effort within the AI software development ecosystem, involving organizations such as OpenAI, Google, Cursor, and Factory, and is now stewarded by the Agentic AI Foundation under the Linux Foundation.

The format functions by utilizing standard Markdown files located at the root of a repository or within nested directories. Coding agents are programmed to parse these files to understand project-specific requirements, such as development commands, dependency management, or linting standards. Because the format is simple and flexible, it avoids proprietary constraints and works across a wide variety of AI coding assistants, including Aider, VS Code Copilot, Devin, and Windsurf. In larger repositories like monorepos, developers can use nested AGENTS.md files to provide granular instructions for sub-projects, with the agent automatically favoring the most locally relevant configuration.

Some of the key features are:

• Predictable Context: Provides a centralized, recognizable location for AI agents to retrieve project-specific operational instructions.
• Universal Compatibility: Compatible with a broad range of leading AI coding agents and development environments.
• Standard Markdown: Uses familiar, human-readable syntax without requiring complex schema definitions or proprietary structures.
• Hierarchical Support: Supports nested files in subdirectories, allowing for specific guidance for different modules within a single repository.
• Living Documentation: Encourages iterative updates to reflect the changing state of a codebase and evolving project needs.

To use AGENTS.md, developers place the file in the root directory of their project or within individual subdirectories for targeted guidance. Once created, the content acts as a programmatic knowledge base that tools can access to perform tasks, run tests, or adhere to specific styling conventions. It acts as a bridge between the existing repository structure and the capabilities of modern AI tools, ensuring that agents can autonomously perform tasks with greater accuracy and less manual intervention.

Some common use cases include:

• Build Automation: Defining clear steps for installing dependencies and starting development servers for an AI agent to execute.
• CI/CD Compliance: Providing instructions on required linting and testing commands that must pass before finalizing a pull request.
• Coding Standards: Communicating strict formatting rules, such as indentation preferences or language-specific patterns, to ensure AI-generated code aligns with project requirements.
• Subproject Configuration: Supplying unique build or testing instructions for specific packages within a large monorepo to avoid configuration conflicts.
• Developer Onboarding: Summarizing complex setup processes or security considerations that a human or an AI agent needs to understand before contributing to the codebase.