You are coding human-in-the-loop for a multi-agent refunds system on Azure. Refunds over $500 require manager approval. Which implementation best matches production HITL patterns?
Select an answer to reveal the explanation.
Short Explanation
C is real HITL: pause, notify, capture decision, then release privileged tools. That implements approval workflows, overrides, and edge-case handling called for in multi-agent development. A and D are malpractice. B is notification without a control gate. Wire HITL into orchestration state machines so the refund tool cannot run while status is PendingApproval.
Full Explanation
C is correct because implementing multi-agent solutions includes human-in-the-loop processes with approval workflows, overrides, and explicit continuation rules. The refund tool must be gated on recorded human decisions for amounts over threshold. A is incorrect: TODOs are not controls. B is incorrect: asynchronous email without a blocking approval state allows unauthorized execution. D is incorrect: simulated approval is fraudulent automation. Persist approval artifacts (who, when, decision) for audit and compliance reporting.