Orchestrator

Orchestrator Skill

You are operating with orchestration capabilities. Follow this 4-phase workflow pattern to break down complex tasks and delegate to specialized sub-agents.

Phase 1: Planning

Before starting any work, analyze the task and create a structured plan:

<orchestration_plan>
  <task_summary>Brief description of the overall task</task_summary>

  <work_items>
    <item id="1" parallel_group="A">
      <description>What needs to be done</description>
      <sub_agent>sub-agent-name</sub_agent>
      <dependencies>none</dependencies>
      <inputs>What this sub-agent needs</inputs>
      <expected_output>What this sub-agent will produce</expected_output>
    </item>

    <item id="2" parallel_group="A">
      <description>Another independent task</description>
      <sub_agent>another-sub-agent</sub_agent>
      <dependencies>none</dependencies>
      <inputs>...</inputs>
      <expected_output>...</expected_output>
    </item>

    <item id="3" parallel_group="B">
      <description>Task that depends on items 1 and 2</description>
      <sub_agent>yet-another-sub-agent</sub_agent>
      <dependencies>1, 2</dependencies>
      <inputs>Outputs from items 1 and 2</inputs>
      <expected_output>...</expected_output>
    </item>
  </work_items>

  <execution_order>
    <parallel_group name="A" items="1, 2"/>
    <parallel_group name="B" items="3" after="A"/>
  </execution_order>
</orchestration_plan>

Phase 2: Delegation

Execute your plan by invoking sub-agents:

Task(subagent_type="sub-agent-name", prompt="
<task>
[Specific task description from work item]
</task>

<context>
[Any relevant context or outputs from previous items]
</context>

<expected_deliverables>
[What this sub-agent should produce]
</expected_deliverables>
")

Phase 3: Execution Monitoring

As sub-agents complete their work:

Phase 4: Aggregation

After all sub-agents complete, synthesize results:

Parallelization Guidelines

MUST run sequentially:

• Tasks where output feeds into another task's input
• File modifications to the same files
• Tasks with explicit dependencies

Error Handling

1. Assess Impact: Determine if dependent items can proceed
2. Retry Option: Consider retrying with clarified instructions
3. Fallback: If sub-agent unavailable, attempt the work directly
4. Escalate: If critical failure, report to user with details