Playbooks
A playbook is a curated, bounded objective — a preset that fills in a procedural focus and a suggested verify command for a common kind of maintenance. They're the fastest way to get reliable runs.
Available playbooks
| Playbook | What it does |
|---|---|
| Improve documentation | READMEs, module docs, doc comments, typos |
| Lint / format | Resolve formatter & linter findings (mechanical, no behaviour change) |
| TODO / FIXME burndown | Resolve one small, self‑contained TODO/FIXME |
| Add a missing test | Add one focused test for an untested function |
| Dependency hygiene | Update one outdated dependency safely (minor/patch only) |
Apply one
In Repo settings → Playbook, pick one. It sets the focus (and suggests a verify command). You can edit either before saving — the playbook is a starting point, not a lock.
Define your own
The built‑ins are starters, not the whole story. In Settings → Playbooks, add your own curated templates for work specific to your stack ("migrate to v2 API", "our lint policy") — a name, the procedural focus the agent runs, and a suggested verify command. Custom playbooks appear in the repo dropdown alongside the built‑ins (which stay read‑only). Edit or delete them anytime.
Why playbooks work
Each playbook focus is procedural: orient briefly, take the first safe win, and commit. That framing keeps agents from over‑exploring and producing no‑op runs. They're also the taxonomy explore mode draws from when it proposes its own work.
Playbooks vs profile presets
A playbook is a one‑time starting point you apply to a repo's settings. A profile preset is a named, schedulable work stream a repo keeps — you can run several side by side (round‑robin or on a schedule) and trigger them from the CLI/API. Seed a preset from a playbook, then schedule it.