Multi-Agent Orchestration Patterns

Design multi-agent systems with orchestrator, peer-to-peer, and hierarchical patterns. Covers task decomposition, agent handoffs, shared state, and coordination so a team of agents stays reliable instead of looping.

This skill architects multi-agent systems. It chooses between orchestrator-worker, peer-to-peer, and hierarchical topologies, decomposes tasks across specialized agents, defines handoff protocols and shared state, adds review and quality gates between steps, and applies guardrails against runaway loops and conflicting actions.

When to use

Use when designing a system where multiple LLM agents collaborate — deciding topology, handoffs, shared state, and coordination so the team is reliable.

Examples

Design an agent team

Pick a coordination topology

Design a multi-agent research system with an orchestrator that delegates to search, analysis, and writing agents, with a review gate before output

Fix agent loops

Add coordination guardrails

My agents hand off in circles and repeat work — add coordination patterns and termination conditions to keep the workflow converging