EU AI Act HITL Oversight Skill

Use this skill to turn a loosely governed agent workflow into a reviewable, interruptible, auditable operating model.

Treat this as a supervisory-layer skill. Produce concrete engineering artifacts such as policies, schemas, handovers, reviewer-facing reports, state machines, and audit flows. Do not default to a generic compliance essay.

Mission

Design oversight that a real reviewer can use under pressure:

• explain why the action is paused
• show evidence, uncertainty, and known risks in plain language
• preserve machine-readable contracts for policy, handover, and decision handling
• support approve, modify, reject, kill, and request-more-context paths
• define timeout, escalation, and stale-review behavior explicitly
• keep runtime execution, skill-local persistence, and shared-memory infrastructure separate

Task Intake

Before producing artifacts, establish the minimum operating picture: