video-use for Claude Code vs PilotCut: A Rendered File or a Timeline You Keep Editing

You already edit code with Claude Code, and you want it to cut your talking-head takes too. Browser Use's video-use skill does that with a folder of footage, ffmpeg and a chat: you drop in raw takes and get a rendered file back. PilotCut takes the same agent and puts it on an editor timeline, so the result is a project you keep editing, by prompt or by hand.

What video-use does

All of this comes from the project's README.

• An open source skill. video-use is MIT licensed. You paste a setup prompt into Claude Code, Codex or another agent with shell access, and the agent installs it.
• Transcribes through ElevenLabs. It needs ffmpeg and an ElevenLabs API key. It makes one ElevenLabs Scribe call per source and packs the word-level transcript into a text file the agent reads.
• Looks at frames on demand. A composite image of filmstrip, waveform and word labels is drawn only at decision points, such as an ambiguous pause.
• Renders an edit decision list. The pipeline goes transcribe, pack, reason, edit decision list, render, self-check. It re-renders up to three times if the check finds an issue.
• Hands back a file. The output is a rendered final.mp4 in an edit folder next to your sources. Session notes carry over to the next session.
• Asks before it cuts. The README says the agent "proposes a strategy, waits for your OK" before it renders.
• Adds finishing passes. It cuts fillers and dead space, color grades, fades audio at every cut, burns subtitles and generates animation overlays.

What changes when the agent works on a timeline

• Adjust one cut, keep the rest. Every agent cut lands on the PilotCut timeline. Move one cut point by hand, and the other cuts stay where they are.
• Read the plan the tool returns. Ask for a cut, a silence pass or filler-word removal, and PilotCut itself returns the list of changes first. Nothing moves until the agent applies it.
• Scroll back through the agent's work. 250 edit history entries show what the agent did, and the agent reads the same history.
• Leave notes the agent respects. Drop a marker for the agent. It can resolve your marker and never deletes it. Each marker carries its origin and a confidence value.
• Cuts that respect breath. Transcript cuts snap to breath-safe points, and the agent proposes each cut from evidence.
• Hand off to Final Cut. The agent can write the timeline as FCPXML 1.11 with absolute media paths.
• Deliver in several shapes. The canvas takes 16:9, 9:16, 1:1, 4:3, 3:4, 4:5 and 21:9 from one project.

How it is built

• Editing tools, not shell scripts. The agent calls 83 purpose-built editing tools. The buttons and the command line call the same tools through the same mutation coordinator, so an agent edit undoes exactly like a click.
• Propose, then commit, on named actions. The timeline cut, the silence, word and dead-air cuts, generated chapters, caption import and patch, and voiceover and camera replacement return a plan on the first call and mutate nothing. A second call applies it. Removing a track applies at once, and so does accepting a B-roll slot.
• The approval step is in the chat panel. The server never waits for a human. The Core Agent chat panel gates each tool on its risk tier. An agent on the CLI or a third-party MCP client is not gated and can send both calls.
• A filmstrip check. The agent can pull a filmstrip of the timeline and look at the frames before it hands the edit back.
• A dry run where it fits. The canvas and layout tools and the effect tool return the result without writing it. Asset and track tools have no dry run.

video-use vs PilotCut for agent editing

Facts

Questions

What is video-use?

An open source skill from Browser Use, MIT licensed, that lets Claude Code, Codex or another agent with shell access edit a folder of footage with ffmpeg and return a rendered file (README).

What does video-use need to run?

ffmpeg and an ElevenLabs API key for transcription. The agent installs the skill and asks for the key once (README).

Can I fix one cut by hand after the agent edits in PilotCut?

Yes. Every agent edit lands on the PilotCut timeline. Move the cut yourself and ask the agent to continue from your version. Both edits sit in the same history.

Does PilotCut work with Claude Code and ChatGPT Codex?

Yes. Claude Code, ChatGPT Codex, Grok Build, GitHub Copilot, OpenCode and Antigravity drive PilotCut over local MCP and the command line, on the plan you already pay for.

Does PilotCut wait for my approval before the agent cuts?

The server never waits for a human. A cut returns its plan first and applies on a second call. In the Core Agent chat panel that second call is gated on the tool's risk tier. An agent on the CLI can send both calls itself.

What does PilotCut cost?

PilotCut is free during beta. There is no published price after the beta. Your agent bills to the plan you already pay for.

Also read

• Mirage Tesseract vs PilotCut
• GroundSlate has no timeline: PilotCut keeps one for you and your agent
• How to use Claude Code or Codex with MCP and CLI for video editing
• Descript filler and silence cuts vs PilotCut propose-then-commit

Conclusion

PilotCut puts Claude Code, ChatGPT Codex and the other agents you already pay for on a timeline you keep editing, with a plan before the cuts that change meaning and a 250-entry history. Free during beta.

Try PilotCut