Playbooks

Playbooks are reusable automation patterns learned from successful runs. When a run completes successfully and merges, it can be promoted into a playbook — a proven recipe that the orchestrator reuses for similar future tasks.

How playbooks work

1. A run succeeds — an agent completes a task (e.g. “add dark mode toggle”) and the PR merges.
2. Promote to playbook — you (or the system automatically) promote the run’s execution pattern into a reusable playbook.
3. Future matching — when the orchestrator decomposes a similar intent, it grounds DAG nodes against matching playbooks instead of planning from scratch.
4. Compounding — over time, your workspace accumulates proven patterns, making execution faster, cheaper, and more reliable.

Creating a playbook

From a successful run

POST /runs/:runId/promote-playbook

Pre-conditions: the run must be in merged state.

The resulting playbook captures the run’s agent sequence, tool calls, and configuration — immediately available for the composer (POST /workflows/compose) to reference.

From the UI

1. Navigate to Automations → Playbooks.
2. Click + Create to build a playbook from scratch, or browse the Agents Window for merged runs and click Promote to Playbook.
3. Edit the playbook’s name, description, and trigger conditions.
4. Publish.

From the entity API

POST /entities
{
  "kind": "playbook",
  "name": "Add Feature Toggle",
  "spec": {
    "trigger": "feature request with toggle pattern",
    "steps": [...],
    "agents": ["dev", "qa"],
    "estimatedCostUsd": 0.35
  }
}

Playbook lifecycle

Versioning

Like all entities, playbooks are versioned:

POST /entities/:id/publish    → create immutable version
GET /entities/:id/versions    → version history
POST /entities/:id/fork       → copy-on-write into a new playbook

Compounding intelligence

Playbooks are the mechanism behind Cendriix’s compounding behavior. The Insights dashboard (Observability → Insights) visualizes:

• How many runs reused existing playbooks vs. planned from scratch
• Cost savings from playbook reuse
• Agent reliability improvements over time

See also: Flows, Agents, Runs & Orchestration