What problem does it solve? Product demo and training videos go stale every time the UI changes, and re-recording them by hand is slow and inconsistent. This Skill turns a version-controlled JSON scene script into the single edit surface: change the script, rerun the capture, and get a fresh video without ever editing the mp4. ## Core Features & Use Cases - Script-driven browser capture: Drives a real browser through scenes with Playwright, draws a pointer, trail, captions and chapter labels via an in-page overlay, and encodes a constant-60fps H.264 mp4 with ffmpeg. - Authenticated journeys: Supports login, consent, bundle granting, HMRC authorisation and VAT return submission steps, with credentials read from environment variables and scanned out of all outputs. - Accessible publishing outputs: Every run produces the mp4 plus a .vtt caption file, a Markdown transcript and per-scene stills, satisfying WCAG SC 1.2.1 for silent videos. - Use Case: After a UI redesign, update the targets in videos/tour.json, run the capture against a local static server with --stills-only to verify every scene, then dispatch the GitHub workflow to record the final video against production. ## Quick Start Ask the AI to record the tour scene script in videos/tour.json against a locally served instance of web/public and review the generated contact sheet of stills.
