What problem does it solve? AI agents routinely describe visual work in words instead of showing it, and ask humans to decide in units they don't think in (pixels, dB, internal ids). This Skill fixes the human-facing exchange: it governs when and how an agent shows finished work, narrates progress, and asks for decisions a person can actually make. ## Core Features & Use Cases - Show, don't describe: Render side-by-side comparisons and mockups for visual or textual decisions instead of asking in raw units, with artifacts opened where the human will actually see them (browser or inline, never a bare file path). - Structured decision requests: Batch multiple open questions into one interactive HTML decision page with radio options, a marked recommendation, and a downloadable JSON answer file that gets harvested back into the work queue. - Progress narration and reporting: Twenty-two rules covering capture echoes, mid-work beats, offline windows, status boards, honest result reporting, and plain-language registers in the product's own vocabulary. - Use Case: After an agent finishes a UI change, it opens a rendered before/after page in the browser with a one-line passport stating what needs the human's word, rather than pasting a wall of test counts and version strings into chat. ## Quick Start Ask the agent to show you the finished change rendered side by side with the previous version and to present any open choices as one decision page with its recommendation marked.
