Public reference
Playbook
Definition
A playbook is a governed, agent-consumable procedure for connecting a specific AI coding agent to CoreStory-grounded context and working against it. Each playbook is written for machines first: preconditions to check, ordered steps to execute, and validation criteria to confirm the agent is actually using retrieved organizational context rather than model priors.
Key properties
- Required sections (per infra#266): Preconditions (what must be true before starting — endpoint reachable, auth configured, project ingested), Procedure (ordered grounding/retrieval steps), and Validation (how to confirm grounded context was used).
- Every playbook page carries the same governance metadata as all pages on this site: owner, last-reviewed date, status, canonical URL, and source (registry: /linking-map.json).
- Concrete instances: Claude (Claude Code / Claude Agent SDK), Cursor, Devin, CoreStory Agent (via CoreStory MCP).
- LLM-first style: bullets, steps, and config blocks — not marketing prose.