How to use the Create AGENTS.md File agent skill

What does the create-agentsmd skill make the agent do?

When the skill is active, the agent follows a set prompt to create a complete AGENTS.md file in the repository root. AGENTS.md is a plain Markdown file that works as a "README for agents": a predictable place to put the setup commands, test instructions, code style rules and pull request conventions that AI coding tools need. It complements README.md rather than replacing it.

The skill tells the agent to inspect the project before writing, looking at files such as package.json, Makefiles, CI configuration and documentation, so the commands it records come from the project itself.

• Places AGENTS.md at the repository root, or at subproject roots for monorepos.
• Uses a template with sections such as Project Overview, Setup Commands, Testing Instructions, Code Style, Build and Deployment, and Pull Request Guidelines.
• Asks the agent to test the commands it lists and to explain why steps are needed.

How do you install the create-agentsmd skill?

The skill ships as a ZIP file from this page. You can install it in one of three ways, depending on how you work with Claude.

For Claude Code, unzip the download into your personal skills folder at ~/.claude/skills/ so it is available in every project. To limit it to one repository, unzip it into that project's .claude/skills/ folder instead.

In the Claude apps, open Settings, go to Capabilities, then Skills, and upload the ZIP there.

• Personal (all projects): ~/.claude/skills/
• Single project: .claude/skills/ inside the repository
• Claude apps: Settings → Capabilities → Skills, then upload the ZIP

How do you trigger the skill with a natural request?

You do not need a special command. Describe the task in plain language, and say that you want an AGENTS.md file for the repository. The agent should recognise the request and use the skill's prompt.

Be specific about scope. If the repository is a monorepo, say so, and say whether you want one root file or one file per package.

• "Create an AGENTS.md file for this repository."
• "Write AGENTS.md for this project so coding agents know how to build and test it."
• "Make an AGENTS.md at the root and in each package of this monorepo."
• "Update the existing AGENTS.md, the test commands have changed."

What should you check in the generated file?

The skill asks the agent to test its commands, but generated instructions can still be wrong. Before you commit the file, run each listed command yourself and confirm it does what the file says. Check that package names, test paths and CI details match your project.

Read the file as the agent that will use it. Remove anything vague, and add any project rule the agent could not have found in the code, such as a review policy or a required lint step.

• Run every command in the Setup and Testing sections.
• Confirm the PR title format and required checks match your team's practice.
• Remove guesses about deployment or secrets that the repository does not actually show.

What about safety, scripts and network access?

According to the skill listing, create-agentsmd requires nothing extra to be listed, does not run scripts and does not use the network. The skill is a prompt that guides the agent's writing. The commands the agent records in AGENTS.md are copied from your project, so they can run on your machine when another agent or person follows them.

Review AGENTS.md before merging it, because other coding agents will read it and may act on what it says.

• The skill itself runs no scripts and makes no network calls, according to its listing.
• Any build, test or deploy command written into AGENTS.md could affect your machine or environment if run, so check it first.
• Do not put secrets or credentials in the file.

Which projects benefit most from this skill?

The skill suits maintainers who want AI coding agents to follow the project's build and test steps, teams starting a new repository who want a consistent instruction file, and monorepo owners who need separate AGENTS.md files per package. It is less suitable where the agent cannot verify commands, because unchecked instructions may mislead.

• Maintainers who want agents to run the right build and test steps.
• Teams starting a new repository with no agent instructions yet.
• Monorepo owners who need per-package AGENTS.md files.