video-use: drop raw footage in a folder, get final.mp4 back
Browser Use's open-source editing skill reads your video as a transcript instead of watching it, then cuts, grades, captions and checks its own render.
video-use is a skill for Claude Code, Codex and similar agents. It had about 29K GitHub stars when we checked on Oct 14 2026. You point the agent at a folder of raw takes, say something like 'edit these into a launch video', and it writes edit/final.mp4 next to your sources. It handles talking heads, tutorials, montages and interviews, with no presets.
The key design choice is that the model never watches the video. ElevenLabs Scribe turns each take into a word-level transcript with speakers and audio events like laughter and applause, and all of it is packed into a file of about 12KB. The agent only asks for a filmstrip-plus-waveform image when it reaches a decision point, such as an ambiguous pause or a choice between two retakes. It's the same idea as giving a browser agent the page's DOM instead of a screenshot.
What it does to the cut: it removes filler words and dead air, color grades each segment, adds 30ms audio fades at every cut, and burns in subtitles (two-word uppercase chunks by default). It can also spawn sub-agents to build animated overlays in HyperFrames, Remotion, Manim or PIL. Before it shows you anything, it inspects every cut boundary of the rendered file and re-renders up to three times if something is off. Session memory is kept in project.md.
Setup: paste the setup prompt from the README into your agent, or clone the repo, symlink it into ~/.claude/skills, run uv sync, install ffmpeg, and add an ELEVENLABS_API_KEY. The agent proposes an editing strategy and waits for your approval before it cuts. A hosted version is waitlist-only.