Claude Code Agent Teams
5 hours ago
- #parallel-work
- #agent-teams
- #claude-code
- Agent teams are experimental and disabled by default; enable them via `CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS`.
- Agent teams allow multiple Claude instances to work together, with a lead coordinating tasks and synthesizing results.
- Use cases include research and review, new modules/features, debugging with competing hypotheses, and cross-layer coordination.
- Agent teams differ from subagents in communication and coordination; teams allow direct messaging between teammates.
- Enable agent teams by setting `CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1` in `settings.json` or environment variables.
- Teams can be displayed in-process (single terminal) or split panes (requires tmux or iTerm2).
- Teammates can be assigned specific roles or models, and plan approval can be required for complex tasks.
- Delegate mode restricts the lead to coordination tasks only, preventing it from implementing tasks directly.
- Teammates communicate via a shared task list and mailbox, with automatic message delivery and idle notifications.
- Best practices include giving teammates enough context, sizing tasks appropriately, and avoiding file conflicts.
- Limitations include no session resumption with in-process teammates, task status lag, and one team per session.