What problem does it solve? AI agents that act on a user's behalf often feel like slot machines: they act without asking, hide what they're doing, can't be stopped cleanly, and silently overwrite outputs. This Skill provides a complete design method for the surfaces of an acting AI so users always know what it will do, what it is doing, and how to stop or reverse it. ## Core Features & Use Cases - Autonomy Gate: Classifies every agent action by reversibility and blast radius to decide between autonomous action with undo, staged review, or a mandatory approval gate. - Run and control surfaces: Checklists for plan display, named-step progress, streaming latency budgets, interrupt/cancel/undo behavior, and receipts for every autonomous action. - Gates, memory, and regeneration: Five-part approval gate anatomy, review queue patterns, memory visibility and dosing rules, and version-rail regenerate affordances. - Use Case: Designing an agent that applies a supplier price list to 46 products and emails affected customers — the Skill produces a full spec covering staged edits, per-recipient send gates, a review queue for outliers, and undo receipts. ## Quick Start Ask the AI to design the plan display, approval flow, progress states, and undo behavior for an agent that performs a multi-step task on behalf of users.
