What problem does it solve? When an AI agent completes a large implementation, the human reviewer cannot realistically read the entire diff, yet a plain summary hides consequential behavior changes, unverified claims, and open product decisions. This Skill turns completed AI work into an inspectable review grounded in actual results and evidence. ## Core Features & Use Cases - Behavior-centered explanation: Reconnects to the original request and explains each changed behavior with before/after results, mechanism, and consequences, organized by what happens rather than by file. - Evidence integrity: Distinguishes observed results, source-derived inferences, supplied evidence, and unverified claims, and never presents mocks or prior AI summaries as proof. - Decision routing: Surfaces at most three unresolved product decisions at a time, keeps confirmed defects separate as blockers, and records a resumable return point when review pauses. - Use Case: After an agent finishes a checkout change, generate a self-contained HTML review page showing actual before/after behavior, flagging a confirmed duplicate-charge defect as a release blocker, and listing only the product questions that genuinely need human judgment. ## Quick Start Ask the AI to use the human-review skill to walk you through the completed change with actual before-and-after results and evidence, highlighting only what needs your decision.
