How does the Supervisor pattern work?

A central supervisor agent analyzes a task, determines which specialized agents are needed, plans their execution order—handling both sequential and parallel steps—and manages the overall workflow.

What is the Fan-Out/Fan-In pattern?

It is a workflow where a single task is distributed among multiple specialized agents simultaneously (fan-out) and their results are then collected and combined (fan-in) into a final output.

Does this skill help with agent communication?

Yes, it includes an Agent Communication Bus pattern that enables message passing and shared state history between different agents to ensure they stay aligned during execution.

How are conflicts between agents resolved?

This skill provides patterns for resolution based on agent confidence scores or by using an LLM as an independent arbitrator to evaluate disagreeing findings and decide on the most likely correct path.

Multi-Agent Orchestration

Name: Multi-Agent Orchestration
Author: yonatangross

byyonatangross

•

データサイエンスとML

Coordinates multiple specialized AI agents using architectural patterns like fan-out/fan-in, supervisors, and synthesis for complex task execution.

概要

This skill provides production-ready patterns for orchestrating complex multi-agent systems within Claude Code. It enables the implementation of sophisticated workflows where tasks are delegated to specialized specialists, results are aggregated through parallel execution, and conflicting outputs are resolved via confidence-based or LLM-arbitrated logic. Ideal for building robust AI systems that require diverse expertise—such as simultaneous security, performance, and architecture reviews—it ensures coherent result synthesis and reliable communication between independent agents through a dedicated message bus.

主な機能

Automated conflict resolution using confidence scores or LLM arbitration
Intelligent result synthesis and executive summary generation
Fan-Out/Fan-In parallel execution patterns
Supervisor-led task routing and dependency management
Agent communication bus for asynchronous message passing
29 GitHub stars

ユースケース

Building complex AI applications that require specialized sub-tasks managed by a central supervisor.
Conducting multi-dimensional code reviews (Security, Performance, and Quality) in parallel.
Aggregating diverse research or data analysis outputs into a single, synthesized report.

概要

主な機能

Automated conflict resolution using confidence scores or LLM arbitration
Intelligent result synthesis and executive summary generation
Fan-Out/Fan-In parallel execution patterns
Supervisor-led task routing and dependency management
Agent communication bus for asynchronous message passing
29 GitHub stars

ユースケース

Building complex AI applications that require specialized sub-tasks managed by a central supervisor.
Conducting multi-dimensional code reviews (Security, Performance, and Quality) in parallel.
Aggregating diverse research or data analysis outputs into a single, synthesized report.