# How to write a playbook

One job per file. A skill teaches an agent how to act. A playbook is a recipe a person runs. This file is the playbook guide.

The id is the filename stem. Never rename an id that already resolves. `attach_skill(name)` and `/playbooks/<id>.md` are the contract.

A new or enriched file is not done until `title`, `summary`, `limitations`, and `how_to_use` are real. Summary is one or two sentences. Limitations say what it will not do. `how_to_use` is the steps, with the real tool or prompt. "Not written yet" is only for old files the page already shows.

Do not add a second copy. Do not invent a playbook category if film, product, edit, or sound fits. Do not nest the 126 skills. The reader walks `playbooks/<category>/<id>.md`. `README.md`, `CLAUDE.md`, and `agent.md` are not playbooks.

Enrich means edit the file in place. Not a new id for a reworded file.

Quality bar: no contradiction of an existing skill. Skill tags come from the existing domain list: image, video, audio, 3d, live-stream, film, story, podcast, briefing, commerce, comic, medical-imaging. A playbook names the skills and caps it uses. No private ops, keys, VM names, or auth. Do not restore the withheld library unless the user names a file.

## How a user does this with their own agent

1. Paste the link to this file into the chat first: `/playbooks/agent.md`.
2. Name the existing id to improve, or the new id and category.
3. The agent edits that one file and fills `summary`, `limitations`, and `how_to_use`.
4. The user accepts only if the limitations are still honest.
5. A batch is one file per change, with ids checked against the catalog.