Instruction Design Principles for sub-agents state that instructions must have:

  1. A narrow scope: The task should be focused and well-defined.
  2. Precise output formats: This allows for predictable aggregation of results by the orchestrator.
  3. Explicit success criteria: This prevents incomplete processing and ensures the sub-agent knows when its task is truly done.

Why it matters: Well-designed instructions are crucial for reliable delegation to sub-agents.

Connections:

Sources: