agent-teams
File System Protocol Coordinates Multi-Claude Agent Teams
Install & Use
Copy this prompt and send it to your AI assistant (Claude / Cursor / TRAE / Codex / WorkBuddy etc.) to auto-install:
Help me install this AI Skill: agent-teams. It is used for: File System Protocol Coordinates Multi-Claude Agent Teams Full Skill content: https://321skill.com/skills/agent-teams-x-10/raw/index.md Read that page and install it.
The prompt includes a link to the full Skill content. You can also view the full content.
This Skill addresses the lack of a unified coordination mechanism when multiple Claude agents work in parallel. In practical development, when needing to handle multiple files, modules, or conduct multi-agent code reviews simultaneously, developers often have to manually manage task distribution and progress synchronization among agents, which is inefficient and error-prone.
Usage is straightforward: you simply define the team name, member list, and roles in the configuration file (~/.claude/teams/<team-name>/config.json), then start each Claude agent via command line specifying its identity. After that, you can use natural language instructions to create teams and assign tasks. Agents communicate via JSON files with fcntl locks, requiring no database or network layer.
It is ideal for development teams needing to implement multiple modules in parallel or conduct large-scale refactoring, especially for agent developers and project managers already using Claude Code who wish to enhance multi-agent collaboration efficiency. It also suits scenarios requiring coordination of multiple AI agents for complex tasks, such as parallel multi-file modifications or cross-subsystem refactoring.
We recommend using this tool when tasks can be independently decomposed and have high parallelism. Note that it is primarily suitable for loosely coupled parallel tasks and not for single tasks requiring tight sequential reasoning. Additionally, ensure tmux and the Claude CLI are installed in the environment, and the experimental flag CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 is enabled.
Key Features
Unlike multi-agent solutions coordinated via APIs or message queues, Agent Teams are entirely based on a pure file system protocol and `fcntl` locks. This requires no additional infrastructure (such as databases, message brokers, or network services), making it suitable for offline or low-network environments and natively supporting localized deployment.
Limitations
Requires `tmux` and the Claude CLI, and the Claude Code experimental flag (`CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1`) must be enabled. Not suitable for single tasks requiring tight sequential reasoning. Communication between agents involves polling, introducing some latency.
FAQ
How do I create a team?
Write the team configuration (including name, member list, and roles) to `~/.claude/teams/<team-name>/config.json` via MCP or direct API.
How do agents communicate with each other?
Each agent has a dedicated inbox file (JSON). Communication uses `fcntl` exclusive locks to prevent concurrent write conflicts, and agents poll their own inboxes to retrieve messages.
Installation guide for AI assistants
If your AI coding assistant (Claude Code, Cursor, TRAE etc.) can see this page, send it this message to auto-install:
Visit https://321skill.com/skills/agent-teams-x-10/raw/index.md to read the original Skill definition (Markdown format) for agent-teams, and install it according to the instructions.
Raw Markdown URL for AI: /skills/agent-teams-x-10/raw/index.md