/scenelens — Claude watches a video, smarter
You don't have a video input; this skill gives you one. Compared to a fixed-fps frame grab, scenelens:
- Picks frames at scene changes — content-aware sampling instead of time-uniform sampling. Same frame budget, far better signal.
- Runs OCR on every frame — on-screen text (slides, code, terminals, dashboards) is extracted as text alongside the image, so you don't burn vision tokens reading static pixels.
- Auto-chunks long audio —
[Description truncada. Veja o README completo no GitHub.]