agent_orchestration

Structure the prompt into four sections: Context (business goal, current state, constraints), Task (primary deliverable, format, scope boundaries), Requirements (must-haves, quality criteria, integration points), and Additional Context. State constraints explicitly and define measurable success criteria.

Multi-agent orchestration routes clarified requests to specialized agents like architects or mentors, each handling a distinct responsibility. Use it when a task spans multiple expertise domains and a single generalist agent would produce shallow or misaligned outputs.

Use it after a clarification phase has resolved ambiguities and before routing to a specialist agent. It is designed for handoffs between agents, not for initial requirement gathering or final implementation.

Run the validation phase checklist: simulate the conversation flow, verify the router directs intents correctly, and audit token usage per step. Define explicit fallback behavior when no agent matches an intent.

Common mistakes include dumping entire conversation histories instead of distilled context, vague task definitions, missing budget or timeline constraints, and omitting what happens after the agent finishes. Each delegation should state its downstream consumer.