2.0 KiB
2.0 KiB
| description | argument-hint | |
|---|---|---|
| Orchestrate multiple agents for complex multi-domain tasks |
|
Agent Dispatch
Coordinate multiple agents to accomplish a complex task requiring different expertise areas.
Instructions
- Consider the recent conversation history, your context, and the task to be accomplished.
- Specific user instructions should be followed unless they are contradictory to the task at hand. $ARGUMENTS
Steps
- Load Skills — Use the Skill tool to load:
- outfitter:subagents — agent routing and orchestration patterns
- outfitter:context-management — for long-running tasks, teaches Task state persistence
- Consider — Ultrathink and analyze the task, consider the complexity, sequence of steps, and agent requirements.
- Planning — Use the Plan subagent to research the codebase and design an orchestration strategy
- Report — Present the orchestration plan (which agents, what sequence, expected handoffs)
- IMPORTANT: After presenting the orchestration plan, proceed directly to execution.
- Do not wait for approval unless the task is high-risk (destructive changes, production deployment, security-sensitive).
- Execute — Dispatch agents according to the plan, passing context between them
- Persist — Update Tasks throughout with agent IDs, decisions, and progress (survives compaction)
Planning Process
Ensure you've loaded the outfitter:subagents skill. Then coordinate with the Plan subagent to design the orchestration plan. Task the Plan subagent to:
- Explore the relevant parts of the codebase
- Identify which roles are needed (coding, reviewing, research, testing, etc.)
- Determine the best available agents for each role
- Design the execution sequence (sequential, parallel, or hybrid)
- Return a concise orchestration plan
After receiving the plan, think about if you agree with it, make adjustments where necessary, and proceed with the next steps mentioned above.