Playbooks
A Playbook is a reusable, outcome-driven recipe for agent-assisted work.
A Playbook packages repeatable work so another agent can pursue the same result with explicit Variables, bounded access, a task graph, and completion proof. It is more than a prompt and more than a checklist: it defines the contract around the work.
The Playbook model
| Part | What it answers |
|---|---|
| Outcome | What user-facing result should this Run make true? |
| Claim | What must be true, and what evidence would prove it? |
| Variable | What information may the agent need? |
| Capacity | What external ability or connector access may the work require? |
| Skill | What reusable operating guidance should the agent load? |
| Milestone | How is the work grouped into meaningful stages? |
| Task | What bounded unit of work can become runnable? |
Source-backed packages
Each Playbook lives in its own Playbook Repository. The canonical package definition is playbook.doku.ts; supporting instructions and page content live beside it. Doku builds and validates that source before it can become a Published Playbook Version.
A minimal repository begins like this:
playbook.doku.ts # typed Playbook definition
doku.config.ts # package kind and stable package identity
README.md # repository notes
instructions/ # optional task instruction files
page/content.md # optional user-facing Playbook page copyDesign from the result backward
Start with the Outcome, not the task list.
- Write the result in language a user would recognize.
- List the Claims that would make the result trustworthy.
- Define the evidence required for each Claim.
- Declare the Variables and Capacities needed to do the work.
- Create Tasks that produce the evidence and satisfy the Claims.
- Group Tasks into Milestones that make progress legible.
This order keeps the Playbook outcome-driven instead of turning it into an unverified sequence of instructions.
Next
Author a Playbook or go deeper on Outcomes and Claims.