Quick Answer: A Claude Code video editor is a set of skills that teaches Claude Code to handle the first pass of an edit in DaVinci Resolve Studio. Claude imports footage, cuts retakes and dead air, places B-roll, writes captions, and helps export. You still make the creative calls. The skills are instructions and rules, not software, and they're licensed CC BY-NC 4.0, so they can't be used commercially.
If you run a YouTube vlog or a class channel, you already know where the hours go. It isn't the creative decisions. It's scrubbing through 40 minutes of footage to find the three usable takes, deleting the false starts, and matching captions to a voice that has drifted into mumbling. A Claude Code video editor takes that mechanical layer off your plate so your time goes into the parts that need taste.
The project is appariciojunior/video-editor-skills, a GitHub repository of twelve skills. Here's what it does, where it gets tricky, and what you need to know before you install it.
What these skills actually are
The most important thing to understand is that this package contains no code. It's a set of Markdown skill files that Claude Code reads. Each skill teaches Claude a method, a set of numbers, and a list of traps the author learned on real edits. When you ask for help, Claude builds the small scripts it needs with you, tailored to your machine and your Resolve project.
That distinction matters for trust. You can read every instruction Claude will follow before it touches a file. You're not installing an opaque binary that edits your footage.
The skills split into a few groups:
- Workflow and setup:
video-editor-setupandedit-video, which run the flow from ingest through to a report. - Cutting decisions:
rough-cutdecides what to keep, such as retakes, false starts, and punchlines.vlog-pacinghandles silence detection, pads, and jump-cut zooms. - Story and sound:
broll,music-and-sound, andvoice-and-audiocover B-roll selection, music cues, ducking, and loudness. - Output:
captions,motion-overlays,figma-roundtrip, andyoutube-export.
How the first-pass workflow runs
Once setup is done, the day-to-day loop is short. Here's the sequence the edit-video skill follows:
- Drop the raw clip into your
raw video/folder. The skill never modifies the original. - Claude transcribes the audio through a speech-to-text API that returns word timestamps.
- It identifies retakes, false starts, and dead air, then proposes cuts.
- It builds a new timeline in your Resolve project, adding clips rather than editing your template.
- It reports back what it cut, what it kept, and what needs your ear.
Step 5 is where the value compounds. You give feedback in plain words, such as "too choppy" or "music louder on B-roll." Claude writes each lesson into your CLAUDE.md, so the next video starts from a better baseline. Over a few videos, the feedback loop does more work than any single prompt.
Where the Resolve scripting gets hard
Here's where most people's expectations break. The cutting logic is the easy part. The hard part is the DaVinci Resolve scripting API, which fails in ways that don't always announce themselves.
The resolve-scripting skill documents what works, what silently fails, and what crashes Resolve outright. A silent failure is the dangerous category. A call returns without error, but the timeline doesn't contain what you asked for. You only find out when you watch the export and a clip is missing.
That's why the package follows a strict rule: Claude only adds new timelines and new clips. It never touches your template or existing projects. If something goes wrong, you lose a timeline you can delete, not a project you spent weeks building.
The common advice says to let AI do the whole edit and review at the end. That's the wrong order. Review each stage as it happens. A bad transcript cut poisons every caption and every music cue that follows it.
Audio, music, and captions
Audio is where a first pass either sounds professional or doesn't. The voice-and-audio skill covers loudness normalization and a recipe for diagnosing a muffled voice, which usually comes down to room tone and EQ rather than the microphone you bought.
For music, the music-and-sound skill pairs Epidemic Sound cues to sections of your video. It also covers a ducking envelope, which lowers the music under speech automatically, and a restrained approach to sound effects. The principle is that music should support the pacing, not compete with the voice.
Captions are the most visible output. The captions skill produces one-line captions, reads them before import, and checks for errors. Reading first matters. A caption that's correct in the transcript but split across lines in a bad place still reads poorly on a phone screen.
| Stage | What Claude handles | What stays with you |
|---|---|---|
| Retakes and dead air | Detects silence, picks the final take | Whether a joke or pause is worth keeping |
| B-roll | Finds and cuts clips by action | Story fit and emotional timing |
| Music | Proposes Epidemic Sound cues per section | Final mood and volume against the voice |
| Captions | Builds and checks one-line captions | Tone, emphasis, and brand wording |
| Export | Applies Deliver settings | Final watch-through before publish |
Motion graphics and the Figma round trip
For brand graphics, the motion-overlays skill uses Remotion, a React-based framework for programmatic video. The skill times lower-thirds and title cards to your spoken words and keeps them clear of the speaker's face and the caption zone. That last constraint is easy to forget, and graphics that sit on top of captions look amateur fast.
The figma-roundtrip skill handles design. You design the brand kit in Figma, and the skill keeps the code and the design boards in sync. This adds a dependency. If you don't already work in Figma, you can skip this layer and use static graphics instead. The rest of the pipeline works without it.
The license trap most people miss
Here's the counter-intuitive part. The skills are free to use, but the license is CC BY-NC 4.0, which means no commercial use. The README is explicit: you can't sell these skills or a modified version, and you can't use them in anything sold, including paid courses, templates, products, or services.
This catches creators who plan to turn their editing workflow into a client service or a paid template pack. Your own YouTube channel is a grey area only if you monetize it through ads and sponsors. The author says to ask directly for commercial use. Check the terms before you build a business on them, not after.
Setup, keys, and safety
Setup takes a single conversation. Say "set up my video editor," and Claude follows video-editor-setup, asks about your camera, Resolve project, and brand, and builds the scripts with you. The package includes templates/config.example.json with starting values.
Two safety rules deserve attention:
- Create the key file yourself. Make
system/keys.envwith one line,SPEECH_TO_TEXT_API_KEY=your-key. Claude is designed never to read or print it, and it must never be committed to version control. - Nothing publishes without you. The workflow stops before publish, so the final watch-through and upload are your decisions.
You'll need DaVinci Resolve Studio, because the free version lacks external scripting. You also need ffmpeg, Python 3.9 or newer, and a speech-to-text API with word-level timestamps. Epidemic Sound MCP, Figma MCP, and Node with Remotion are optional.
Frequently Asked Questions
What is a Claude Code video editor?
A Claude Code video editor is a set of skill files that teach Claude Code to perform the first pass of a video edit in DaVinci Resolve. It handles transcript cuts, silence removal, B-roll, captions, and export settings. The skills contain no executable code. Claude reads them and builds the scripts it needs for your setup.
How do I automate a first-pass video edit with Claude?
Install the skills, then ask Claude to set up your video editor. Create a system/keys.env file with your speech-to-text key yourself, drop footage into your raw folder, and open your Resolve project. Then say "edit the new video." Give feedback in plain words, and Claude records each lesson in your CLAUDE.md.
Do I need DaVinci Resolve Studio to use these skills?
Yes. The workflow depends on Resolve's external scripting interface, which is available in Studio but not in the free version. You also need ffmpeg and Python 3.9 or newer.
Can I use a Claude Code video editor for paid client work?
Not under this project's license. The package is CC BY-NC 4.0, which prohibits commercial use, including selling the skills or using them in paid services. Ask the author for commercial permission before you build a business on them.
Why does the Resolve scripting fail silently?
Some Resolve API calls return without an error but don't produce the change you requested. The resolve-scripting skill documents these cases. Because the failure is silent, you should check each stage of the timeline by eye, not trust a clean exit code.
Start small: pick your next vlog, run only the cutting stage, and watch the result before you add captions or music. Then note what Claude got wrong and tell it in plain words. Try this on one real video this week, and pass this to anyone spending evenings scrubbing footage for usable takes.