explainer

Provide the repo root URL and the skill activates GitHub mode: it scans the README, detects a live demo URL, authors an 8-10 beat script, records the browser walkthrough, and composites an avatar lipsync into a 1280x800 video.

Pass any https URL and the generic-URL mode runs a pre-flight check for bot blocks and cookie banners, then records a beat-sheet walkthrough of the page with element-targeted zooms and narrated TTS audio.

Yes, pass --avatar with a hosted image URL and --voice with a MiniMax voice ID to override the identity-store defaults. Without flags, the skill silently resolves your registered avatar and voice, generating a default presenter if none exists.

Pika is the default, rendering in about 2-5 minutes with more dramatic head motion. Kling produces tighter face-centered presenter shots but takes 5-30 minutes and stalls past roughly 90 seconds of audio, so it suits high-stakes renders.

The generic-URL pre-flight aborts when it detects captcha or bot-block pages such as Cloudflare challenges or access-denied responses, since headless Chrome would record the block page instead of real content.

The beat sheet targets 65-80 seconds and 165-200 spoken words. Audio over 90 seconds is rejected because the Kling lipsync stage stalls beyond that ceiling, and narration under 30 seconds triggers a TTS retry or abort.