What is an AGENTS.md file?

AGENTS.md is a Markdown file at the root of a repository that gives AI coding agents project instructions: setup and test commands, code style, project structure and boundaries. It is an open convention supported by many coding agents, so check that yours reads it.

What is the difference between AGENTS.md, CLAUDE.md and .cursorrules?

They hold the same kind of instructions for different tools. CLAUDE.md is read by Claude Code, .cursorrules is the older Cursor format that Cursor now replaces with Project Rules in .cursor/rules, and AGENTS.md is a tool-neutral convention. Check each tool’s documentation for the files it reads.

Can I use one file for all coding agents?

Often yes. Keep AGENTS.md as the main file, then reuse it in CLAUDE.md with an @AGENTS.md import line, or copy the text. Test the result in each tool, because support differs.

What should I put in AGENTS.md?

Short, specific instructions: the tech stack, exact build, test and lint commands, code style rules, project structure, testing and pull request rules, and clear boundaries for what to always do, ask first about, and never do. Leave out anything an agent can read from the code.

How long should an agent instructions file be?

Keep it short. The file is loaded into the agent’s context, so long files use space and bury important rules. As a rule of thumb, stay under about 150 lines and write each rule so it can be checked.

Is it safe to build my agent file here?

Yes. The builder runs in your browser and nothing is uploaded. Do not add keys or passwords, because these files are usually committed to the repository. The Share link button puts your text in the URL, so only share it with people you trust.

AGENTS.md Builder: AI Coding Agent Instructions

AGENTS.md / CLAUDE.md / .cursorrules builder

Save on Google

Free AGENTS.md generator. Build AGENTS.md, CLAUDE.md, .cursorrules or a Cursor .mdc rule for your coding agent, with checks and a share link. Everything runs in your browser. Nothing is uploaded.

Configure agent file

Output file

Project nameWhat the project does (one or two sentences)Tech stack (comma or one per line)

Commands

InstallDev serverBuildTestLint

Rules

Code style (one per line)Project structure (path: purpose, one per line)Testing (one per line)Git and pull requests (one per line)

Boundaries

Always do (one per line)Ask first before (one per line)Never do (one per line)

Results & Output

What is an agent instructions file?

Coding agents such as Claude Code, Cursor and others read a Markdown file in your repository before they work. A good file tells the agent how to build and test the project, which style to follow, and what it must never touch. This builder writes AGENTS.md, CLAUDE.md, .cursorrules or a Cursor .mdc rule from one form, with checks for missing commands, vague rules and secrets. Tools differ in which files they read, so check your tool’s documentation.

FAQ

Question	Answer
What is an AGENTS.md file?	AGENTS.md is a Markdown file at the root of a repository that gives AI coding agents project instructions: setup and test commands, code style, project structure and boundaries. It is an open convention supported by many coding agents, so check that yours reads it.
What is the difference between AGENTS.md, CLAUDE.md and .cursorrules?	They hold the same kind of instructions for different tools. CLAUDE.md is read by Claude Code, .cursorrules is the older Cursor format that Cursor now replaces with Project Rules in .cursor/rules, and AGENTS.md is a tool-neutral convention. Check each tool’s documentation for the files it reads.
Can I use one file for all coding agents?	Often yes. Keep AGENTS.md as the main file, then reuse it in CLAUDE.md with an @AGENTS.md import line, or copy the text. Test the result in each tool, because support differs.
What should I put in AGENTS.md?	Short, specific instructions: the tech stack, exact build, test and lint commands, code style rules, project structure, testing and pull request rules, and clear boundaries for what to always do, ask first about, and never do. Leave out anything an agent can read from the code.
How long should an agent instructions file be?	Keep it short. The file is loaded into the agent’s context, so long files use space and bury important rules. As a rule of thumb, stay under about 150 lines and write each rule so it can be checked.
Is it safe to build my agent file here?	Yes. The builder runs in your browser and nothing is uploaded. Do not add keys or passwords, because these files are usually committed to the repository. The Share link button puts your text in the URL, so only share it with people you trust.