About Video Production
Video work splits into steps that an agent can run: write the script, build or cut the footage, add captions, render, and check the result. The skills here each own one of those steps, so a single request does not have to carry the whole pipeline.
Two approaches dominate. Code-driven video uses Remotion, where scenes are React components and a render turns them into a file. Footage-driven video uses FFmpeg and Whisper to cut long recordings into vertical clips with burned-in captions. Analysis skills read a video through its transcript first and look at frames only when visuals matter.
When choosing, start from the source material. Data, product screens and text suit Remotion. Talking-head recordings and podcasts suit the clip pipelines. Generated drama and ads suit the prompt-writing skills, which need a storyboard and a target model.
Check what each skill needs on your machine, such as a Remotion license for commercial use, local Whisper, FFmpeg or API keys, and whether it renders at once or waits for your approval first.