video-use is an agent skill that edits video through a coding agent such as Claude Code, Codex, Hermes, or OpenClaw. You drop raw footage into a folder, ask the agent to edit it, and it inventories the sources, proposes a strategy, waits for your approval, and renders edit/final.mp4 next to your sources. It handles talking heads, montages, tutorials, travel footage, and interviews without presets.
The model never watches the video; it reads it. One ElevenLabs Scribe call per source produces word-level timestamps, speaker diarization, and audio events, packed into a single takes_packed.md of roughly 12 KB. When a decision needs visual context, such as an ambiguous pause or a retake comparison, a timeline_view tool renders a filmstrip, waveform, and word-label PNG for a time range. This replaces frame dumping (tens of millions of tokens) with a small text file plus a few images.
The pipeline runs transcribe, pack, reason, edit decision list, render, and self-evaluate. The self-eval step runs timeline_view on the rendered output at every cut boundary to catch visual jumps, audio pops, and hidden subtitles, and re-renders up to three times before showing you a preview. Cuts follow speech boundaries and silence gaps, and the skill applies 12 production rules while leaving creative choices open.
Features
- Filler removal: cuts umm, uh, false starts, and dead space between takes
- Color grading: applies a grade per segment (warm cinematic, neutral punch, or a custom ffmpeg chain)
- Audio fades: 30 ms fades at every cut
- Subtitles: burns in styled captions, two-word uppercase chunks by default
- Animation overlays: generates overlays through HyperFrames, Remotion, Manim, or PIL, each in a parallel sub-agent
- Self-evaluation: inspects the render at each cut boundary before presenting it
- Session memory: persists context in project.md so later sessions resume where you left off
- Agent-agnostic: registers as a skill with any agent that has shell access, and can run always-on through Browser Use Box
