agent-teams
Coordinates Claude Agent Team Collaboration via Filesystem Protocol
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: Coordinates Claude Agent Team Collaboration via Filesystem Protocol Full Skill content: https://321skill.com/skills/agent-teams-x-6/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 AI Agents collaborate. In practical development, when multiple Claude instances need to process tasks in parallel, it often requires manual session management and state synchronization, leading to inefficiency and a high potential for errors. Agent Teams enables loosely coupled team collaboration by having each Agent run independently in a tmux pane, coordinated through a filesystem protocol (JSON files + fcntl locks).
Usage is straightforward: simply create a team configuration file (located at ~/.claude/teams/<team-name>/config.json), then launch multiple Claude CLI processes via tmux, specifying parameters like team name and identity for each instance. You can then use natural language instructions to assign tasks and monitor progress. The Agents automatically communicate and coordinate via an inbox message queue and task files, requiring no database or network layer.
It is particularly well-suited for teams or individuals needing to handle large-scale code refactoring, multi-module development, or cross-file code review in parallel. This is especially valuable for developers already using Claude Code who wish to enhance collaboration efficiency, as well as data analysts or operations engineers who need to run multiple independent analysis tasks concurrently.
We recommend using this tool in scenarios involving multiple independent subtasks that benefit from loosely coupled collaboration. Note that it primarily relies on tmux and the filesystem, making it unsuitable for single-file modifications or tasks requiring tight, sequential reasoning. Also, ensure all Agents use the same Claude version and model, and have the experimental environment variable CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 enabled.
Key Features
Unlike LangChain's AgentExecutor, it requires no database, daemon processes, or network services. It achieves lightweight coordination solely through the filesystem (JSON + fcntl locks), avoiding dependencies on additional infrastructure. This makes it particularly suitable for rapidly launching multi-Agent collaboration in local development environments.
Limitations
Requires Claude Code CLI and tmux to be installed, and the experimental environment variable `CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1` must be enabled. It does not support Windows systems (due to tmux dependency). Communication between Agents relies on polling, introducing some latency.
FAQ
What dependencies are required to install this Skill?
You need to install Claude Code CLI and tmux first. Then, install this Skill via the Night Market tool (conjure) and set the environment variable `CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1`.
How can I view the task progress of each Agent in the team?
Agent Teams stores task files in the `~/.claude/tasks/<team-name>/` directory. Each task exists as a JSON file with an incrementing number. You can directly inspect these files or query the current status through the Lead Agent's session.
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-6/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-6/raw/index.md