TextAdvanced

Define when an automation must stop and hand off to a human

Max Submitted by Max Added 12 days ago
You are designing a human-in-the-loop policy for an automation that can change external state. Return: 1. DECISION MAP - task step | available evidence | reversible or irreversible | impact | automation authority. 2. ESCALATION RULES - explicit triggers for ambiguity, low confidence, missing permission, policy conflict, unusual value, repeated failure, user distress, sensitive data, and irreversible actions. 3. HANDOFF PACKET - exact context the human receives: objective, evidence, attempted actions, current state, risk, options, and recommended next step. 4. CONTROL BEHAVIOR - pause, retry, compensate, cancel, timeout, and safe default for each trigger. 5. AUDIT LOG - event fields needed to reconstruct who or what decided, acted, approved, and changed state. 6. TEST SCENARIOS - normal, borderline, and adversarial cases with expected handoff behavior. Rules: confidence alone cannot authorize a prohibited action. Never hide failed attempts. Preserve idempotency and require fresh approval after material state changes. WORKFLOW: {{describe}} TOOLS AND SIDE EFFECTS: {{list}} USERS AND ROLES: {{list}} POLICIES AND LIMITS: {{paste}}

How to use it

Review thresholds with operations, security, and the people receiving escalations. Test both the handoff and the resume path.

Compatible popular AI tools

These tools are mapped to this prompt based on their capabilities.