Guide · Agent skill
How to use the video-use agent skill: install and edit video
Updated · written with AI from the project's own documentation and checked against our catalogue data
video-use is an agent skill that lets an AI agent edit your video by conversation. You describe the edit in plain language, the agent transcribes the speech, proposes a plan, and renders the result with ffmpeg. This guide explains what the skill does, how to install it, how to ask for edits, and what it needs before it works.
Video Usebrowser-use/video-use · MIT Download the skill Quick start
- 1Download and unzip the skill. Download the ZIP from this page and unzip it into ~/.claude/skills/ for all projects, or into a project's .claude/skills/ folder for one project.
- 2Install the prerequisites. Make sure ffmpeg and ffprobe are on your PATH, and that Node.js 22+ is available if you plan to use HyperFrames animations.
- 3Install Python dependencies. Run uv sync in the video-use repo, or pip install -e . if you use pip.
- 4Add your ElevenLabs API key. Set ELEVENLABS_API_KEY in your environment or in a .env file at the video-use repo root.
- 5Put footage in a folder. Copy your raw video files into one folder. The source files are left untouched; outputs go to an edit subfolder.
- 6Open your agent and describe the edit. Start your agent in that folder and describe the result you want in plain language, such as a length, subtitles, or a tone.
- 7Approve the plan and review the render. Confirm the plain-English plan the agent proposes, then review preview and final outputs and ask for changes.
What the video-use skill makes the agent do
Once installed, the agent takes your raw video files and transcribes the speech word by word using the ElevenLabs Scribe API. It works out cut points from the transcript and from silences between speech, then asks you to confirm a plain-English plan before it changes anything.
After you approve the plan, the agent renders the edit with ffmpeg. The skill describes color grading, optional animated overlays, and burned-in subtitles as part of that render. It saves its notes and outputs into an edit folder next to your footage.
- Cuts out filler words and dead space, based on the transcript and silences
- Applies color grading to segments
- Adds animation overlays when you ask for them
- Burns subtitles into the video
- Saves outputs and working notes in an edit folder beside your footage
Install the video-use skill from the ZIP
Download the ZIP of the skill from this page and unzip it. You can place it in your user skills folder, which makes it available in every project, or in a project's own .claude/skills/ folder, which limits it to that project.
If you use the Claude apps instead, you can upload the skill under Settings → Capabilities → Skills.
- User-wide: unzip into ~/.claude/skills/
- Per project: unzip into the project's .claude/skills/ folder
- Claude apps: Settings → Capabilities → Skills, then upload the skill
- The skill folder is named video-use, and its entry file is SKILL.md
Finish setup: dependencies and the API key
The install guide covers cloning the repo, installing dependencies, ffmpeg, skill registration, and the API key. Run the dependency step inside the video-use repo. Setup is needed once; the agent checks it at the start of later sessions.
The ElevenLabs key is read from the environment or from a .env file at the video-use repo root. If the key is missing, the agent asks you to paste one and writes it to that .env file, not to your footage folder.
- ffmpeg and ffprobe must be on your PATH
- Python dependencies installed with uv or pip
- ELEVENLABS_API_KEY set in the environment or in .env at the repo root
- Node.js 22+ only if you want HyperFrames animation slots
- yt-dlp, HyperFrames, Remotion and Manim are installed only when first needed
uv sync # or: pip install -e .Trigger the skill with a natural request
You do not need special syntax. Put your raw footage in a folder, open your agent there, and describe what you want. The skill asks questions about the material, such as content type, target length, pacing, subtitles, and must-keep or must-cut moments, then proposes a strategy.
Stay in the loop at the confirmation step. The skill is designed to wait for your approval of the plan before it cuts anything, so you can correct the approach early.
- "Edit these clips into a 60-second launch video with burned-in subtitles."
- "Cut the filler words and the long pauses from this interview."
- "Make a talking-head video with a warm grade and a short title overlay at the start."
- "Turn these tutorial takes into one clean video, keeping the second take of step three."
How a session runs
The skill works in a loop: ask, confirm, execute, iterate, persist. It transcribes once per source and caches the transcript, so it does not re-transcribe unless the file changes. It reads a packed phrase-level transcript as its main view and looks at frames only at decision points.
Outputs go into an edit folder inside your videos directory, including the edit decision file, the master subtitle file, intermediate clips, and the final render. The skill is designed never to write inside its own project directory.
- Inventory: inspect the sources and transcribe them
- Converse: the agent asks questions shaped by your material
- Propose: a short plan you approve before any cut
- Execute and iterate: render, review, and adjust
- Persist: notes saved to a project memory file in the edit folder
Safety notes: scripts, network and your footage
The skill runs scripts and uses the network. Audio from your footage is sent to ElevenLabs for transcription, so do not use the skill on sensitive recordings you are not willing to upload to that service. Optional tools may be installed the first time they are needed.
Animation work is built inside its own slot folder, not at the repo root. Before you run the skill on private material, read the install guide and the SKILL.md file so you know which scripts and services it will call.
- Audio is sent to ElevenLabs Scribe for transcription
- Keep your API key in .env at the repo root, not in your footage folder
- Optional tools (yt-dlp, HyperFrames, Remotion, Manim) install on first use
- Review the plan before approving it, since execution follows it
What people use it for
- Cutting filler words and dead air from a talking-head recording
- Turning interview footage into a tighter edit with subtitles
- Producing a short social video with burned-in captions from existing recordings
- Assembling a tutorial from multiple takes and keeping the best one
- Adding a simple animated overlay or title to a travel montage
Frequently asked questions
How do I install the video-use skill?
Download the ZIP of the skill from this page and unzip it into ~/.claude/skills/ for all projects, or into a project's .claude/skills/ folder. In the Claude apps, you can instead upload it under Settings → Capabilities → Skills. You still need the dependencies and the ElevenLabs API key for it to work.
What does video-use require?
It needs ffmpeg and ffprobe on your PATH, Python with uv or pip to install dependencies, and an ELEVENLABS_API_KEY for transcription. Node.js 22+ is needed only for HyperFrames animation slots. yt-dlp, HyperFrames, Remotion and Manim are installed on first use if needed.
Is video-use free?
The project is released under the MIT licence. The ElevenLabs API used for transcription is a separate service that needs your own API key, so check its terms for any costs. This guide does not state prices.
Does video-use send my videos to the cloud?
Audio from your footage is sent to ElevenLabs for word-by-word transcription. The skill's documentation says that outputs are kept in an edit folder beside your footage. If your recordings are sensitive, do not use this skill on them.
Can I use video-use without a point-and-click editor?
Yes, that is its design. You work by describing edits in conversation. It has no menus or presets, so if you need a visual timeline or a fixed template library, a point-and-click editor will suit you better.
Why won't the agent cut my video right away?
The skill is built to confirm a plain-English plan before touching the cut. Approve the plan or ask for changes, and the agent will then execute and render the edit.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
video-use suits creators, marketers, and developers who are comfortable with ffmpeg, Python, and API keys and want to describe edits in plain language. If you need a point-and-click editor, fixed templates, or you cannot send audio to a third-party transcription service, pick a different tool.
Download the skill