- Home
- Skills
- Image & Video
- Video Editing Kit for Mac
More screenshots
Works with the AI Agents you already use
Video Editing Kit for Mac
Local video-editing toolkit for macOS: split talking-head takes from B-roll, transcribe in nearly 100 languages, log clips into a spreadsheet, sync cameras and check renders. No paid API.
$9.99
Video Editing Kit for Mac
Example session with this skill installed
Here's yesterday's card from the mall shoot (~/Movies/kit-demo/SHOOT_DAY1). Find the talking takes, make a contact sheet and log everything into a spreadsheet for the client.
- Read your context and instructions
- Compiled the video editing kit
- Generated the data export
Done: 16 clips checked, 1 talking take, spreadsheet ready
Nothing in your card folder was moved or deleted. Everything new is next to it, in _AROLL/, _DECUPA/ and the .xlsx.
1. Talking takes (aroll SHOOT_DAY1)
| Clip | Result |
|---|---|
| IMG_4505.MOV | 1 take, 00:00 to 00:18: the model's closing line to camera, in Portuguese ("E vocês já sentiram o cheiro..."). Exported trimmed to _AROLL/ |
| 8 clips (IMG_4461, 4463, 4465, 4468, 4469, 4474, 4480, 4512) | Only short off-camera direction ("Vai lá!", "É isso, vai mexendo aí"), never long enough to be a take, so not exported |
| 7 clips (IMG_4459, 4462, 4476, 4481, 4491, 4499, 4502) | B-roll, no speech |
The language was detected on its own (pt). The director's short cues from behind the camera were too short to count as takes, so only the real line to camera was exported.
2. Contact sheet (contact-sheet SHOOT_DAY1)
SHOOT_DAY1/_contact_sheet.jpg: one frame per clip with its name, 16 clips on one page.
3. Logging spreadsheet (logsheet)
I looked at every clip's frames and wrote the description fields, then built the spreadsheet ("Mall walk-through, spring campaign") with a thumbnail per clip:
| Usability | Clips |
|---|---|
| High | 9 (bag and knit details, pleated dress, rings, cafe, closing to camera) |
| Medium | 5 (window shopping, rack browsing, jewelry counter, short earring shot, escalator) |
| Discard | 2 (IMG_4481 and IMG_4491 are under half a second, accidental record taps) |
Groups: Window shopping · Knit & bag store · Pleated dress · Jewelry counter · Fine jewelry · Perfume · Cafe · Closing to camera · Escalator.
Worth a look before sending to the client: recognizable faces in IMG_4499, 4502 and 4505; passers-by in IMG_4474; IMG_4463 has someone crossing the lens at 00:08.
Next steps
- Drop the IMG_4505 take on the timeline as the closing.
- Want the speech of IMG_4505 as subtitles? "Transcribe the _AROLL folder" gives you .txt and .srt.
- Need the B-roll laid over marked ranges in Resolve? Mark them and say "fill the markers with these folders".
_contact_sheet.jpg
CSV · data export
Example file from a real run - the skill writes it into your workspace.
Connects securely to your tools. The creator never sees your data.
About this skill
Ten terminal commands built during real editing work, plus a Claude Code manual so you can just ask in plain language: "split the takes in this folder", "transcribe this video", "make a contact sheet of the card".
Everything runs on your own Mac with ffmpeg, whisper.cpp and macOS Vision. No paid API, no credits, no upload of your footage.
What you get
- aroll: finds the spoken takes in a folder of raw clips, trims each one with 1 s of handles and names it from what was said. Works by voice and pauses in any language, and also understands slate words and mistakes in English, Spanish and Portuguese ("rolling", "action", "let me start over").
- transcribe and transcribe-playlist: .txt and .srt from a file, a folder or a link, in nearly 100 languages with automatic detection.
- logsheet: logs a folder of clips into an .xlsx with thumbnails, metadata and speech (sheet labels in English, Portuguese or Spanish). Claude can fill in the shot descriptions for you, in your language.
- contact-sheet: one frame per clip with the file name, to check a card at a glance.
- find-offline: scans a render for DaVinci Resolve's "Media Offline" frame before you deliver.
- sync-cameras: syncs a second camera to the main one by audio, and can build the synced track in Resolve.
- fill-markers: fills Resolve timeline markers with B-roll overlays from matching folders.
- merge-dupes: merges duplicated folders ("SELECT", "SELECT 2") into one.
- get-audio: extracts the audio of a video link as MP3 (only use with content you have the rights to).
- 4 LUTs (.cube + .dctl with intensity slider): Stadium Warm and three film-inspired looks.
- Tested ffmpeg recipes in the manual: stabilization, render checks, side-by-side comparisons, noise cleanup.
Requirements
- macOS only. Apple Silicon recommended; Intel works through MacPorts.
- About 2 GB of disk for the speech models (downloaded once at install).
- DaVinci Resolve Studio only for
sync-cameras --resolveandfill-markers. The LUTs work in free Resolve. - Claude Code is optional: the commands also work straight from the terminal.
- Docs in English, Portuguese and Spanish.
Install
Unzip the kit and ask Claude Code: "install this kit". The installer shows what it will do and asks before changing your ~/.zshrc. zsh install.sh --uninstall removes only what it added.
macOS only. DaVinci Resolve Studio (paid) is needed only for sync-cameras --resolve and fill-markers; every other command and the LUTs work without it. The speech models take 1.6 GB of disk. aroll is tuned for talking-head footage where the speaker fills a good part of the frame; wide interview shots need --audio-only. Single slate words ("Sorry.", "Cut.") are only recognized when whisper punctuates them as a short sentence; long phrases like "let me start over" are the reliable way to mark a mistake. Shot descriptions in logsheet use your own agent's tokens (that step is optional).
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 11 days ago
Requires
Permissions
Allowed Hosts
Everything runs locally; no footage is uploaded and no paid API is called. Network is used only to install tools (Homebrew or MacPorts), download the speech models once from huggingface.co, and, when you ask for it, fetch the audio of a video link with yt-dlp (it reads your Chrome cookies for YouTube). The installer shows its plan and asks before adding a marked block to ~/.zshrc (with backup); --uninstall removes only what it added. Commands write their output next to your media (_AROLL, _DECUPA) and never delete or move your files, except merge-dupes, which merges folders on purpose.