agentic-ui-reviewlisted

# Agentic interface review Checks that a user can understand what an agent is about to do, stop it, approve the risky parts, see what happened, and recover when it goes wrong. The bar is calibrated trust: the user trusts the agent exactly as much as its reliability deserves, no more and no less. Conversation copy and turn design belong to `conversational-ux`; accessibility mechanics to `better-accessibility`; dark patterns in consent or upsell steps to `dark-pattern-review`. This skill owns the lifecycle of an agent action as the user experiences it. ## Inputs - The interface: screens, prototype, or the components that render agent plans, tool calls, results and errors. - What the agent can do: the tool list with each tool's blast radius (read, write, send, pay, delete) and reversibility. - The autonomy model: what runs without asking, what asks, who can change that. ## Workflow ### 1. Map the action lifecycle For each agent capability, trace the four moments the user meets it: **before** (intent stated, plan shown, permission asked), **during** (progress, current step, ability to pause or stop), **after** (what changed, where, how to undo), **failure** (what the agent knows, what it tried, what the user can do). Note where any moment is silent. Silence at "before" on a write or send action is the most common `HIGH` finding. ### 2. Check the pattern set Read [references/patterns.md](references/patterns.md) for each pattern with the guideline it derives from and