multi-agent-coordination-playbook

Instructions

Use this skill when you need to choose a multi-agent coordination architecture (or decide how to evolve your current one) after deciding that multiple agents are appropriate.

1) Start with the simplest pattern that could work

• Prefer a simple coordination pattern first.
• Observe where it struggles in your real workload.
• Evolve toward a more complex pattern only when the failure mode is clear.

2) Choose a pattern based on structure (not “sophistication”)

Use these quick-fit rules:

• Generator-verifier: choose when output quality is critical and you can make evaluation criteria explicit.
• Orchestrator-subagent: choose when task decomposition is clear and subtasks are bounded and mostly independent.
• Agent teams: choose when subtasks are independent and benefit from long-running, multi-step work with persistent worker context.
• Message bus: choose when work is event-driven, the workflow structure varies by event, and you expect your agent ecosystem to grow.
• Shared state: choose when work is collaborative and agents need to build on each other’s findings continuously.