multi-agent-patterns

To avoid a supervisor bottleneck in multi-agent systems, choose between supervisor, swarm, or hierarchical topologies based on your coordination needs. This skill provides structured topology guidance and ready-to-use coordination logic to distribute tasks effectively.

Inter-agent communication and task decomposition are implemented using structured state management and explicit handoff protocols. This skill provides Python logic for message passing, weighted consensus voting, and circuit-breaker failure handling to ensure context isolation across specialized agents.

Consensus voting for aggregated multi-agent results uses a weighted confidence mechanism to evaluate outputs from specialized agents. This skill provides Python coordination utilities that implement weighted voting to prevent sycophantic consensus and ensure reliable research outcome aggregation.

This multi-agent orchestration approach supports coordination patterns compatible with frameworks like LangGraph and AutoGen. It manages structured state management and explicit handoff protocols required by these frameworks to ensure context isolation and operational reliability during execution flows.

You should avoid hierarchical multi-agent topologies when your workflow risks lossy information degradation during handoffs or requires decentralized coordination. Instead, use swarm patterns for peer-to-peer coordination or supervisor topologies for simpler, centralized task assignment without complex hierarchy.