What problem does it solve? When an AI agent hits a policy limit, an irreversible action, or an unresolvable data conflict, continuing to chat or dumping a raw transcript on a human operator wastes time and creates ambiguity. This Skill defines a strict escalation protocol: the agent calls escalate_to_human, stops generating prose, and emits a typed payload the operator can act on immediately. ## Core Features & Use Cases - Escalation Preconditions: Detects the three canonical triggers for handoff — limit exceeded, irreversible action, and data conflict — rather than escalating on generic uncertainty. - Typed Handoff Payload: Emits a fixed five-field JSON payload (customer_id, issue_summary, actions_taken, escalation_reason, recommended_action) that is fully self-contained. - End-State Loop Termination: Treats the handoff as the end of the agent loop, enforceable via a PostToolUse hook that closes the session after escalation. - Use Case: A customer requests a USD 2,400 refund but the agent's tier-2 limit is USD 1,000. The agent detects the limit breach, calls escalate_to_human with the complete case payload, and terminates the loop instead of continuing the conversation. ## Quick Start Escalate this refund case to a human operator using the typed handoff payload because the amount exceeds the agent's policy limit.
