Does Agent Teams require an external database?

No, Agent Teams uses a local filesystem-based protocol with JSON files and fcntl locks, ensuring high performance and data integrity without external infrastructure.

What happens if one agent in the team crashes?

The health monitoring module tracks heartbeats. If an agent stops responding, its tasks are automatically released after a 90-second timeout for other agents to claim.

Can I define different roles for my agents?

Yes, the skill supports roles like Architect, Implementer, Researcher, and Reviewer, each with specific capability matrices and risk-level permissions.

What is required to run multiple agents simultaneously?

You need the Claude Code CLI and tmux installed. The skill uses tmux to manage split-pane environments where each agent operates in its own process.

How does the system prevent agents from overwriting each other?

The protocol uses a task ownership model and file locking. For additional safety, it also supports git worktree isolation to keep agent file changes in separate environments.

Agent Teams Orchestrator

Name: Agent Teams Orchestrator
Author: athola

byathola

•

209

•

コラボレーションツール

Coordinates multiple Claude CLI processes through a filesystem-based protocol to enable parallel task execution and multi-agent collaboration.

Agent Teams provides a sophisticated framework for orchestrating multiple Claude CLI processes, allowing them to work together on complex development projects. By utilizing a filesystem-based coordination protocol with fcntl locking and tmux-managed panes, it enables agents to communicate asynchronously and manage shared task queues without external dependencies. This skill is ideal for large-scale refactors, parallelized feature implementation, and automated code review loops, where specialized agent roles—such as Architects and Reviewers—can operate concurrently while maintaining strict dependency management and health monitoring.

主な機能

01Real-time health monitoring with stage-based stall detection and automatic task reassignment.

02Filesystem-based coordination using JSON inboxes and fcntl locking for atomic operations.

03Seamless tmux integration for managing multiple Claude processes in a single terminal session.

04Automated task dependency management with support for parallel agent execution.

05209 GitHub stars

06Role-based agent system including Architect, Implementer, Researcher, and Reviewer archetypes.

ユースケース

01Establishing a multi-agent review loop where one agent implements code and another performs validation.

02Executing complex refactoring tasks that require simultaneous changes across decoupled directories.

03Coordinating parallel implementation of multiple modules or subsystems in a large codebase.

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add athola/claude-night-market agent-teams

For use in Claude.ai and ChatGPT

Download Skill