You are writing explicit escalation criteria with few-shot examples for the system prompt. Which two of the following cases should the agent escalate with escalate_to_human rather than try to resolve itself? Select two.

Answer: A and C — A customer's first message says "I want to talk to a real person," even though the question is a routine shipping-status check the agent could answer in seconds. / A customer asks the agent to match a lower price from a competitor's website; your price-adjustment policy covers only price drops on your own site and says nothing about competitors.

An explicit request for a human should be honored immediately, without first attempting investigation (a), and a policy that is silent on the specific request, such as competitor price matching when policy covers only your own site, is a policy gap that calls for escalation (c) (task 5.2). Frustration alone (b) is not a trigger: acknowledge it and resolve a straightforward case that's within the agent's authority, escalating only if the customer repeats a request for a human. Self-reported confidence (d) is an unreliable proxy for case complexity, and a standard in-window return is squarely within scope.