browser-use / video-use

The package ships two skills, not one: video-use itself, which transcribes footage word by word through ElevenLabs Scribe, proposes cuts and color grades, burns subtitles, and renders through ffmpeg; and manim-video, an unrelated skill for building 3Blue1Brown-style math and algorithm animations with the Manim library. video-use's design principle is to read the transcript rather than the frames — the same structured-data-over-screenshot idea browser-use built its name on for web pages, applied here to video.

This is for someone editing real footage — talking heads, interviews, travel, tutorials — from inside an agent's shell, not a no-code editor. Setup means cloning the repository, symlinking it into your agent's skills directory, running uv sync, installing ffmpeg, and pasting an ElevenLabs key when the agent asks for one; every edit after that costs at least one ElevenLabs transcription call on top of whatever the agent session itself costs.

READ THE FULL ANALYSIS＋

Neither skill has been verified to run. Both come back unverified in our dependency check, which means only that we have not confirmed the setup steps ourselves end to end, not that they are broken.

The README sells a paid product next to the free one. Browser Use Cloud and Browser Use Box, hosted and always-on versions of the same editing flow, are linked from the top of the same page as the open-source skill; the free version is real and self-contained, but it shares its pitch with its own upsell.