Human agents who receive cases from escalate_to_human report that they spend the first several minutes re-asking customers for their account details and what has already been tried. The human queue has no access to the agent's conversation transcript. What change best addresses this?

Answer: A — Have the agent pass a structured handoff in escalate_to_human: verified customer ID, order, root cause found so far, refund amount at issue, and a recommended next action.

Mid-process escalation needs a structured handoff protocol: customer details, root cause, amounts, and a recommended action, so a human without the transcript can act straight away (task 1.4). A raw transcript dump (b) makes the human dig through verbose tool output and contradicts the premise that they work from a handoff, not the transcript. A customer-written summary (c) drops the verified facts the agent already established and shifts the work onto the customer. Escalating before investigating (d) lowers first-contact resolution and gives the human less information, not more.