claude-team

What This Skill Does

The claude-team skill is a high-level orchestration layer for Claude Code, designed specifically for power users who need to manage multiple AI agents simultaneously. By integrating directly with iTerm2’s Python API and the claude-team MCP server, this skill allows you to spawn, monitor, and coordinate isolated Claude Code instances. Each worker operates in its own iTerm2 pane, allowing you to visually oversee development tasks, manage git worktrees per worker, and ensure that your primary context remains clean while sub-tasks are performed in parallel.

Installation

To install this skill, run the following command in your terminal:

clawhub install openclaw/skills/skills/jalehman/claude-team

Ensure that you have completed the prerequisite setup: macOS with iTerm2 installed and the Python API enabled (via Preferences → General → Magic → Enable Python API). You must also have your ~/.claude.json configured with the necessary MCP server credentials for claude-team.

Use Cases

1. Parallel Feature Development: Assign multiple workers to different git worktrees to build independent features or fix separate bugs simultaneously without context switching.
2. Large-Scale Refactoring: Use one worker for the core architectural change and others for updating dependent services or writing necessary unit tests in parallel.
3. Contextual Isolation: Keep your main agent session focused on high-level planning while secondary workers handle the granular execution of specific beads issues.

Example Prompts

1. "OpenClaw, use claude-team to spawn two workers: one to fix the auth bug in project_path ~/repos/api and another to verify the database schema changes for bead cp-202."
2. "Check the status of all current workers and message 'Groucho' to inform him that he should prioritize the unit test suite after finishing the primary task."
3. "List all active workers. Once the 'busy' ones are finished, please close the 'ready' sessions to clean up my iTerm2 layout."

Tips & Limitations

• Safety First: As noted in the core rules, never attempt to make code changes directly through the orchestrator. Always spawn a worker. This ensures that every edit is tracked within a specific git worktree, making rollbacks and history management significantly safer.
• Layout Management: Use the layout parameter in spawn_workers wisely. The "auto" setting is excellent for daily workflows, but use "new" if you are beginning a distinct, large-scale project phase to prevent window clutter.
• Monitoring: Always monitor worker status via list_workers before firing off new instructions to ensure you don't inadvertently overload your system's resources by spawning too many concurrent LLM sessions.