agent-teams
Coordinating Claude Multi-Agent Teams via a 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: Coordinating Claude Multi-Agent Teams via a Filesystem Protocol Full Skill content: https://321skill.com/skills/agent-teams-x-9/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 CLI instances work in parallel. In real-world development, developers often need to run several Claude processes simultaneously to handle different subtasks, but direct communication is difficult, leading to potential data conflicts and missed tasks. Traditional solutions require setting up databases or network services, whereas Agent Teams achieve lightweight coordination through a pure filesystem protocol.
Usage is straightforward: simply create a team configuration file in the terminal, then launch multiple subprocesses using parameters like claude --agent-id, with each process corresponding to an agent. The system automatically creates task queues and inboxes in the ~/.claude/teams/ directory, using JSON files with fcntl locks to ensure concurrency safety. Agents poll their own inboxes to retrieve tasks and write results.
It is particularly suitable for teams or individuals needing to process multiple related tasks in parallel, especially developers already using the Claude CLI for coding who want to enhance efficiency in large-scale projects. For example, multiple agents can simultaneously perform code reviews, write unit tests, and refactor modules, exchanging intermediate results via files.
This tool is recommended for large projects with well-defined parallel subtasks. Note that it primarily relies on tmux and the filesystem, making it unsuitable for tightly coupled, real-time collaboration scenarios. Each agent requires a separate Claude CLI process, which consumes significant resources; it is advisable to allocate concurrency appropriately based on machine performance.
Key Features
Unlike sequentially executing tasks within a single Claude session, Agent Teams enable multiple Claude processes to run independently and exchange messages asynchronously via a filesystem protocol, supporting true parallel execution without relying on databases or network layers, resulting in a more lightweight deployment.
Limitations
Requires installation of tmux and Claude CLI, along with enabling the `CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1` environment variable. It is not suitable for tightly coupled, real-time inference tasks, and each agent occupies a separate Claude process, leading to relatively high resource consumption.
FAQ
Does this Skill require installing additional databases or network services?
No, it only depends on tmux and Claude CLI, communicating via the filesystem without the need for databases or network layers.
How do I launch a team and assign tasks?
First, create a team configuration file, then start multiple subprocesses using parameters like `claude --agent-id` and `--team-name`, with each process corresponding to an agent. They will automatically fetch tasks from the task queue and write results.
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-9/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-9/raw/index.md