claude-team
par jalehman
Coding Agents & IDEsRésumé
Orchestrate multiple Claude Code workers via iTerm2.
Description
Ce que fait ce skill
MCP server that spawns and manages teams of Claude Code worker sessions inside iTerm2. Each worker gets its own terminal pane and an optional git worktree, and can be assigned a beads issue to work autonomously. Workers run in parallel while the coordinator's context stays clean.
Each worker runs in a real, visible Claude Code session with isolated context and a dedicated git worktree, so parallel work never pollutes the coordinator's context window and can be inspected or taken over at any time.
Quand l'utiliser
- Fanning out three feature branches to separate workers simultaneously
- Assigning individual beads issues to workers for autonomous parallel implementation
- Monitoring multiple Claude Code agents from a single coordinator terminal
- Waiting for all workers to finish before reviewing and merging their branches
- Sending mid-task corrections to a specific worker without interrupting others
Exemple de workflow
User asks: implement three separate beads issues in parallel using workers
- 1
Calls spawn_workers with three configs, each specifying a bead ID, annotation, and use_worktree true
- 2
Each worker creates a git worktree, runs bd show on its assigned issue, and marks it in_progress
- 3
Calls wait_idle_workers with mode all and a timeout to block until all three complete
- 4
Calls read_worker_logs for each worker to review conversation history and output
- 5
Calls close_workers to terminate sessions, then merges or cherry-picks each worker branch
Three beads issues implemented on isolated branches, ready for review and merge to main
Prérequis
- iTerm2 installed on macOS with Python API enabled (Preferences → General → Magic → Enable Python API)
- mcporter binary installed and available in PATH
Installation
clawhub install jalehman/claude-teamExécute la commande CLI, ou colle l'instruction dans Claude Code, Cursor, Codex ou l'agent de code que tu utilises.