subagent-result-integration

To delegate asynchronous tasks with sessions_spawn, structure the task for the subagent, spawn the session, and track execution using labels and run IDs to avoid premature responses while waiting for completion.

Subagent delegation should be used for agent orchestration when tasks are slow, heavy, or parallelizable, requiring a separate specialized model to process before the parent agent synthesizes the final output.

Result integration works by tracking the child session via runId and childSessionKey, distinguishing between task acceptance and completion, and synthesizing the structured output only after the subagent finishes.

Your parent agent responds prematurely because it misinterprets the subagent status, confusing task acceptance with completion; proper execution tracking via labels and run IDs prevents this premature response.

Yes, you can use sessions_spawn for fan-out orchestration to delegate parallel tasks across multiple subagents, ensuring accurate synthesis of each child agent output before generating the final response.

The best way to track subagent execution is implementing traceability with label, runId, and childSessionKey, ensuring robust auditing and correct identification of subagent results for final synthesis.