What is the 'Fan-out/Fan-in' pattern?

It is a coordination model where a central coordinator decomposes a large task into parts for specialist agents, then merges all their individual results into a single final output.

How does Cowork Sessions handle agent failures during a task?

If an agent fails, the session captures the error and allows other agents to continue. The final output includes partial results and specifically highlights which subtasks succeeded or failed.

Does this skill track the cost of AI model usage?

Yes, it tracks total input and output tokens across all agents involved in a session and provides an estimated cost based on the specific Claude models used (Opus, Sonnet, or Haiku).

Can I run multiple agents at the same time?

Yes, Cowork Sessions supports parallel execution for independent tasks, allowing up to 10 agents to run simultaneously depending on the marketplace item configuration.

Cowork Session Orchestrator

Name: Cowork Session Orchestrator
Author: markus41

bymarkus41

•

생산성 및 워크플로

Orchestrates and manages multi-agent parallel workflows to execute complex tasks through coordinated plugin agents.

Cowork Sessions provides the domain knowledge necessary to launch, monitor, and manage sophisticated multi-agent workflows within the Claude environment. By coordinating multiple plugin agents in parallel or sequential patterns—such as fan-out/fan-in architectures—this skill enables Claude to break down complex projects into subtasks, track resource usage across multiple models, and maintain session state through pause/resume functionality. It is an essential tool for developers needing to automate multi-stage workflows that require specialized agents working in tandem to deliver comprehensive results.

주요 기능

01Real-time resource tracking for tokens, duration, and estimated costs

02Comprehensive session lifecycle management including pause, resume, and cancel

03Robust error handling that captures partial results from failed subtasks

04Multi-agent orchestration supporting Sequential, Parallel, and Fan-out models

059 GitHub stars

06Automated subtask decomposition and specialist agent matching

사용 사례

01Coordinating multi-stage documentation pipelines with different specialist agents

02Parallelizing security audits and performance reviews across a large codebase

03Managing complex code generation tasks that require simultaneous unit test writing

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add markus41/claude cowork-sessions

For use in Claude.ai and ChatGPT