agent-teams

This Skill addresses the lack of a unified coordination mechanism when multiple Claude agents collaborate. In practical development, when multiple agents need to work in parallel on different modules or tasks, developers often face pain points such as chaotic communication, task conflicts, and difficulty in synchronizing progress. Traditional coordination solutions based on networks or databases are too heavy and unsuitable for fast-paced local development scenarios.

Usage is straightforward: simply start a main agent via the claude CLI in your terminal, then use preset team configuration commands to create members and assign tasks. Each member agent runs in a separate tmux pane, communicating and synchronizing tasks via JSON files (with fcntl locks). You can then coordinate the entire team using natural language instructions, e.g., "Have the frontend agent implement the login page, the backend agent prepare the API, and merge when done."

It is ideal for teams or individuals working on parallel development, multi-agent code review, large-scale refactoring, or naturally decomposable tasks. This is especially useful for developers already using Claude Code who want to boost productivity, or for scenarios requiring simultaneous handling of multiple files/modules. Project managers can also have the agent team execute tasks in parallel after requirements are broken down.

We recommend using this tool in a Unix/Linux environment and ensuring tmux and Claude Code are properly installed. Note that it is primarily suitable for loosely coupled tasks; for work requiring tight sequential reasoning (e.g., continuous dialogue), a single agent may be more efficient. Additionally, cross-machine coordination is not currently supported—all agents run on the same machine.