How to use the EGM Bootstrap README Generator agent skill

What is an agent skill?

An agent skill is a folder containing a SKILL.md file and optional supporting files. The SKILL.md file gives an AI agent instructions for one kind of task. When the skill is loaded, the agent follows its workflow instead of improvising from scratch.

egm-bootstrap is one such skill. Its folder includes SKILL.md, agents/openai.yaml, and references/templates.md and references/worker-return.schema.json.

What does the skill make the agent do?

The skill focuses on documentation for a directory tree. Leaf folders that contain source code get a Component Package README (the source calls these Peanuts). Parent folders get a Repository Blueprint that rolls up child interface contracts into a higher-level index (called Hay).

Each generated component README includes an Overview, Interface, Invariants, Side Effects, and a Verification section that lists the test command for that package.

• Discovers the tree breadth-first, one directory level at a time, without recursive search commands.
• Skips hidden and build folders such as .git, node_modules, venv, dist, build, and third_party, and respects .gitignore and .egmignore.
• Detects test commands from manifests, for example npm test for package.json, cargo test for Cargo.toml, go test ./... for go.mod, and pytest for Python projects.
• Skips directories that contain only test code or no source code files.
• Keeps its own status and log files in worker-specific folders outside the repository.

How do you install the skill?

Download the ZIP for this skill from this page and unzip it. Then place the folder in one of these locations, depending on how you use Claude:

For a personal setup, unzip it into ~/.claude/skills/ so it is available across your projects. For a single project, unzip it into the project's .claude/skills/ folder and commit or share that folder as your team prefers.

In the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.

• Personal: ~/.claude/skills/egm-bootstrap/
• Project: .claude/skills/egm-bootstrap/ inside the repository

How do you trigger the skill with a natural request?

The SKILL.md file sets disable-model-invocation to true, so the agent is not meant to start this workflow on its own. Ask for it directly and name the folder you want covered. Be clear about scope, because the skill is designed for subtrees of a repository.

Good requests name the skill or the EGM format and give a path, such as the packages under services/ or a single package you want documented.

• Use the egm-bootstrap skill to write README contracts for everything under packages/billing.
• Bootstrap EGM README.md files for the subtree at services/search, and show me the list of directories first.
• Create EGM.md files instead of README.md for libs/, so our hand-written READMEs stay untouched.

What does the agent need to run it?

The skill needs an agent harness with shell, file-read, file-write, and file-edit tools. Subagent support is recommended for parallel workers. Without it, the skill runs the scout steps inline. A project markdown formatter such as prettier is optional and is only used if the repository already configures one.

The workflow assumes you are working inside a code repository with source files, manifests, and test commands. It does not run network calls, according to the skill's listing.

• Shell, file-read, file-write, and file-edit tools in the agent harness
• Optional: subagent support for parallel workers
• Optional: a configured formatter such as prettier

What are the safety notes?

The skill declares that it runs scripts and does not use the network. It is restricted to documentation work. The agent should not change functional source code, should not run git commit or git add, and should not touch files outside the documentation it generates.

The skill can overwrite README.md. If your repository has hand-written READMEs, set target_filename to an alternate name such as EGM.md. If the discovered tree has more than 15 subdirectories, the agent must stop and ask you for approval before continuing.

Review all changes before you commit them, since the skill leaves version control to you.

• Do not commit until you have reviewed the diff.
• Use EGM.md when existing READMEs must not change.
• Expect a pause and approval prompt on large trees.

What tips help the results?

Start with a single package or a small subtree to see the output format before running a full tree. Prefer a subtree with a clear boundary so the Blueprint index roll-up makes sense.

If a repository has no recognized manifest, the verification command is left as the placeholder <test-command> for you to fill in.

• Run on a small folder first and inspect the generated sections.
• Check the Verification commands against your real CI setup.
• For small projects with a handful of folders, writing READMEs by hand may be faster.