Does this skill support the native Agent Teams feature?

Yes, it specifically includes patterns for CC 2.1.33+ Agent Teams, including mesh topology, peer-to-peer messaging via SendMessage, and shared task list management.

Is using multiple agents more expensive?

Yes, multi-agent workflows typically consume more tokens because each agent maintains its own context. However, this is often offset by the higher quality of results and reduced manual rework for complex tasks.

How does this skill handle conflicting agent outputs?

It provides two primary methods: confidence-based resolution, where the agent with the highest self-reported confidence wins, and LLM arbitration, where a separate model reviews the conflict and decides on the best path forward.

What is multi-agent orchestration in Claude Code?

It is a set of architectural patterns that allows Claude to spawn, manage, and synthesize results from multiple specialized subagents to handle tasks too complex for a single agent session.

When should I use a Supervisor pattern vs. a Fan-Out pattern?

Use Fan-Out for independent tasks that can run in parallel, such as checking a file for three different types of bugs. Use the Supervisor pattern for workflows that require planning, routing, and sequential dependencies.

Multi-Agent Orchestration Claude Code Skill

This skill provides production-ready orchestration patterns for Claude Code, enabling the management of multiple specialized agents working in parallel or sequence. It includes robust implementations for the Supervisor pattern, conflict resolution strategies, and result synthesis, while providing full support for native CC Agent Teams (v2.1.33+) featuring peer-to-peer messaging and shared task lists. It is designed for complex engineering workflows that require diverse domain expertise—such as security audits, architecture reviews, or full-stack feature implementation—where a single agent session might lack the required context depth or specialized focus.

主な機能

01Native CC Agent Teams support for mesh-topology peer communication

02Automated synthesis patterns to merge multiple specialist outputs into one report

0369 GitHub stars

04Fan-Out/Fan-In patterns for high-speed parallel agent processing

05Intelligent Supervisor routing for dynamic task delegation based on intent

06Conflict resolution logic using confidence scores or LLM arbitration

ユースケース

01Large-scale codebase refactoring where changes must be synchronized across disparate modules

02Conducting multi-dimensional code reviews (Security, Performance, Quality) in parallel

03Building full-stack features by spawning and coordinating specialized frontend and backend agents

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add yonatangross/orchestkit multi-agent-orchestration

For use in Claude.ai and ChatGPT

Download Skill