Playbooks

A Playbook is the source of truth for expected agent behavior.

Good Playbooks include:

• actors and roles
• allowed actions
• forbidden actions
• policy thresholds
• escalation rules
• evidence requirements
• known failure modes

Create a draft

wendell playbook create \
  --name "Refund Agent Regression" \
  --workflow-summary "A support agent evaluates refund requests and escalation rules." \
  --source ./refund-playbook.md \
  --project-ref refund-agent \
  --domain customer-support-refunds \
  --extract

Review extracted primitives

wendell playbook review wdraft_abc123

The review output shows extracted sections such as actors, tools, policies, allowed actions, and risks.

It may also include required questions. Required questions must be answered before approval.

Apply review answers

{
  "operations": [
    {
      "op": "answer_question",
      "section": "roles_and_actors",
      "answer": ["customer", "support agent", "manager"]
    }
  ]
}

wendell playbook apply wdraft_abc123 \
  --file ./review.json

Supported review operations

Operation	Purpose
answer_question	Resolve a required Playbook question.
add_primitive	Add a value to a summary section.
update_primitive	Replace a summary section value.
remove_primitive	Remove a summary section value.
reject_primitive	Remove a summary value during review.
resolve_conflict	Record a reviewer decision for a conflict.
accept_assumption	Record approval of a generated assumption.

Approve and generate

wendell playbook approve wdraft_abc123 \
  --reviewer "you@example.com" \
  --generate-suite

Approval fails if required questions remain unresolved.

Publish

wendell suites publish \
  --draft wdraft_abc123

Publishing creates a versioned hosted test suite that can be run with wendell run.