agent-teams
Orchestrating Claude Multi-Agent Team Collaboration 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: Orchestrating Claude Multi-Agent Team Collaboration via a Filesystem Protocol Full Skill content: https://321skill.com/skills/agent-teams-x-5/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 processes work in parallel. In practical development, a single Claude session is limited by its context window and reasoning depth, making it difficult to simultaneously handle tasks that require parallel progress—such as large-scale refactoring, multi-module development, or code reviews—while manually managing multiple terminal windows often leads to task conflicts and progress confusion.
Usage is straightforward: you simply define a team configuration file (containing member roles, identity tags, and colors) and then start the team with a natural language command. The Skill automatically creates a tmux multi-pane session, with each Claude instance running independently. Communication occurs via JSON files in the ~/.claude/teams/ directory, using fcntl file locks to ensure concurrency safety. All tasks are written with incremental numbering into the tasks/ subdirectory. You can then assign tasks with commands like "Have the frontend agent implement the login module while the backend agent concurrently writes the API." Each agent works asynchronously and writes its results.
It is particularly well-suited for teams or individuals engaged in parallel development, multi-agent code review, large-scale refactoring, or cross-module collaboration. This tool is especially valuable for developers who already use the Claude CLI for daily coding but want to overcome the concurrency bottleneck of a single process, as well as for project managers who need to handle multiple independent subtasks simultaneously.
We recommend prioritizing this tool when tasks have natural parallelism (e.g., modifying multiple files, developing independent modules) and avoiding it for small tasks that require tight sequential reasoning, as that would introduce unnecessary overhead. Note that it depends on tmux and the Claude CLI, requires setting the environment variable CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1, and—due to file I/O latency—is not suitable for high-frequency, real-time communication scenarios.
Key Features
Unlike manually switching contexts within a single Claude process, Agent Teams enable loosely coupled multi-process collaboration via a filesystem protocol. This approach supports high-concurrency task assignment without requiring a database or network layer, resulting in a lighter-weight configuration.
Limitations
Depends on a tmux environment and the Claude CLI; requires enabling an experimental environment variable. The file-lock mechanism may become a performance bottleneck under high-frequency writes. Does not support distributed collaboration across machines.
FAQ
How do I start a team?
First, create a team configuration (a JSON file saved to `~/.claude/teams/<team-name>/config.json`). Then, start each agent process using parameters like `claude --agent-id`. It is recommended to use tmux to manage the multiple sessions.
How do agents communicate with each other?
Agents communicate by passing messages via JSON files in `~/.claude/teams/<team-name>/inboxes/`. Atomic writes are ensured using `fcntl` file locks, and each agent polls its own inbox.
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-5/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-5/raw/index.md