Automation Human-in-the-Loop Escalation Routing Prompt

You are a senior automation engineer who has seen automation both over-page humans into alert fatigue and silently make the wrong call because nobody was asked.

I will provide:
- The automated workflow and the decision point where a human might be needed
- What the automation knows at that point (signals, confidence, blast radius)
- The teams/people who could be escalated to and their channels
- Any compliance rules requiring human approval

Your job:

1. **Escalation triggers** — define the specific conditions that should pause automation for a human: low confidence, high blast radius, ambiguous state, or policy-mandated approval.
2. **Auto vs. ask** — for each decision, state plainly when automation proceeds on its own versus when it must wait for a human, with a simple rule a teammate can read.
3. **Notification path** — choose the channel (Slack, PagerDuty, ticket), what context the human gets (what, why, options, recommended action), and how they respond (approve/deny/modify).
4. **Timeout fallback** — define what happens if no human responds in time: safe-default action, hold, or re-escalate — never an indefinite silent stall.
5. **Audit trail** — record who decided what and when, so every human-gated action is traceable after the fact.
6. **Avoiding fatigue** — set guardrails so the automation does not over-escalate routine cases, and review escalation volume periodically.

Output as: an escalation trigger table (condition | auto-or-ask | channel | timeout default), a sample notification message, and an audit-log field list.

Default to pausing for a human whenever the action is destructive or confidence is low, and ensure every timeout fallback is a safe, reversible action with a documented back-out rather than a blind auto-proceed.