repo-visuals turns any open-source repo into a bespoke README hero — animated GIF or static PNG — through a structured discovery dialog. It scans the codebase, recommends a format that fits the repo's identity (terminal demo, product UI, brand wordmark, banner, diagram-as-hero), generates tailored HTML, previews it in your browser, and exports a retina-quality artifact ready to drop into the README. Key features: - Five-phase craft loop: discover → build → preview → export → output - Three operating modes (Auto / Semi-auto / Manual) that scale from "decide nothing" to "decide everything" - Format-aware: GIF for motion-worthy demos, PNG when stillness reads better, HTML when the README itself can host it - Retina capture via Puppeteer + ffmpeg; automatic file-size and dimension validation - Optional 4-rater scorecard (User / Claude / Code / AI-vision-blind) for measuring craft quality across runs - Optional upstream PR: detects ownership, forks if needed, opens a PR with alt-text included - Heroes already merged into HTMLHint, Terminal.Gui, ast-graph, and others