Playbook

Playbook

Also called: 操作手册 · Runbook

A complete procedure for a class of task, written for an agent to follow — usually with decision branches and handling rules, more complete than a single Skill.

Playbook: handle refund1. verify order2. if amount > $200 -> human approve3. else issue refund4. notify customer(branches + decision rules)
Schematic (simplified)

Characteristics

The term is borrowed from operations and support: it is not just linear steps but spells out "if situation X, do this," "when to escalate to a human," and "which preconditions must hold." A playbook for an agent is typically natural language plus explicit branches, keeping the agent inside a constrained decision space rather than fully improvising.

Relation to Skill / workflow

Roughly: a Skill emphasises packaging method and knowledge, a workflow emphasises a code-fixed execution path, and a playbook sits between — prose that walks an agent through the full procedure for a class of situation, including exceptions. A Skill can contain a playbook.

Common misconceptions

  • A playbook does not mean hard-coding the process — it keeps decision points but makes the decision rules explicit too.
  • Steps alone are not enough; a good playbook must cover exception paths and escalation conditions.

Related terms

Sources

  1. Google SRE Book — ch. 14, on runbooks/playbooks (procedure design)
  2. Anthropic — Building effective agents (2024)

Compiled 2026-08-29 · This glossary is compiled from public papers, official specifications, and common industry definitions, and is updated as the field evolves. Corrections welcome.