creating-agent-teamslisted

# Creating Agent Teams ## Overview This skill helps you decide whether a task needs a single agent, parallel subagents, or a coordinated team - and compose effective teams with the right models, agent types, and communication patterns. **Core principle:** Don't create a team when a single sonnet agent could do it in one pass. Teams add coordination overhead - use them only when parallelism or specialization provides real benefit. **Official documentation:** [Agent Teams on code.claude.com](https://code.claude.com/docs/en/agent-teams) ## Prerequisites Agent teams are experimental and disabled by default. Ensure this is enabled before creating any team: ```json settings.json { "env": { "CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1" } } ``` ## Agent Roster Predefined agents the Orchestrator picks from based on task needs. Only spawn agents whose scope is actually touched. | Agent | Model | Type | Role | Default Scope | |-------|-------|------|------|---------------| | **Orchestrator** | opus | general-purpose | Plans, delegates, synthesizes. Never implements directly. | Full project awareness via Explorer | | **Explorer** | haiku | Explore | Answers Orchestrator's questions about the codebase. Finds files, patterns, conventions. | Targeted dirs only — search what's asked | | **Frontend** | sonnet | general-purpose | React components, views, styling, state management | `frontend/` only | | **Backend** | sonnet | general-purpose | Rails controllers, models, services