Auto Edit Workspace skill
A workspace for talking-head video editing: an agent cuts flubs, adds subtitles, cards and sound while you refine in a browser.
- 55 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The agent runs programs, installs tools and writes project files.
- Video and voice are processed locally; external generators receive only what you enable.
Install
- Works in
Manual install
git clone https://github.com/dean9703111/auto-edit-workspace.gitThe first README method. Then open the folder with the agent and say the environment check phrase.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The agent runs programs, installs tools and writes project files.
- Video and voice are processed locally; external generators receive only what you enable.
This is third-party code. Review the repository files before installing.
What it does
Auto Edit Workspace bundles nine skills for editing talking-head video. You put footage, voice, screen recordings and images in the project and then ask the agent in plain language to cut flubbed takes, align subtitles to a script, and add cards, animations and sound effects. The agent pauses at the cut list, framing, storyboard and preview so you can approve each step. Conversion, transcription and rendering run on your own computer, and rendering is built on HyperFrames. A browser editor shows a timeline where you fix subtitles and card text and queue redo requests. The output is an MP4 or layers for CapCut. Rules, templates and sound tables are plain text files, so the next episode inherits your edits. Documentation is in Traditional Chinese.
Who it is for. Creators of educational and expert videos who shoot to camera and want a clean edit without manual cutting.
Good fit when
- A lesson was recorded with flubs and bad takes must be removed.
- Subtitles must match the script rather than the transcript.
- The video needs info cards and scene animations.
- The edit must continue in CapCut with ready-made layers.
Not a fit when
- You want to edit a music video or narrative film.
- The computer cannot handle local transcription and rendering.
- You need documentation in Russian or English.
Example request
Check my editing environment, then cut flubs in projects/example and show the edit list for approval.Limitations
git, Node.js 22, ffmpeg, whisper-cli, auto-editor and HyperFrames are required. The setup skill checks them and installs with your consent. Windows is marked untested in the README. Background images and music use external generation, which is optional and may be paid. Documentation is in Traditional Chinese, though commands and skill names read without translation. Review every proposed step: the agent changes project files and runs installs.
How to disable. Delete the project folder along with `.agents/skills`. Installed ffmpeg, whisper-cli and auto-editor are removed separately through your package manager.
FAQ
How do I get started?
Clone the repository, open the folder with the agent, say the environment check phrase and then 「剪乾淨」 on the example in `projects/example`.
Which agents does it work with?
The README names Claude Code, Codex, MiniMax Code and Cursor.
What must be installed?
Git, Node.js 22, ffmpeg, whisper-cli, auto-editor and HyperFrames. The setup skill checks the list and asks for consent.
Can I edit the result by hand?
Yes. A browser editor fixes subtitles and card text and queues redo requests for the agent.
Where do my files go?
Conversion, transcription and rendering run on your computer. Only the optional image and music generators are external.
Discussion
No questions or reviews yet.
Related
SkillsSkills, a CLI and an MCP server that let Claude Code or Codex make video from a brief: editing, graphics, voiceover, subtitles
A free video editor inside your agent: drop in a raw clip, say "edit this" and get a cut with captions, graphics and sound.
A video editing skill for Claude Code and Codex: the agent cuts pauses and filler words, grades color, burns subtitles and adds animated overlays
chengfeng-videocut
chengfeng-videocut-skills
An agent plugin for editing talking-head video: removing verbal slips, subtitles, on-screen graphics and rendering the final cut
Thirteen HyperFrames video skills: script, motion samples, A/B shot review, voiceover, captions and final checks
A Codex and Claude Code plugin with a local storyboard table: the agent writes shots and makes images, motion and voice-over
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| Auto Edit Workspace (this card) | no | 55 | |
| Video Editing Skill | no | 209 | a general video editing skill |
| HyperFrames | no | 83,283 | the render foundation without a ready editing workflow |
README in short
The workspace holds skills for rough cut, storyboard, style, motion, review and delivery. The agent stops at checkpoints and rules live in text files.
README badge
Original SKILL.md
--- name: auto-edit-setup description: 檢測並安裝這個工作區要用的工具（git、Node.js 22、ffmpeg／ffprobe、whisper-cli、auto-editor、根目錄 npm install 的 HyperFrames、hyperframes 與 auto-editor 兩組外部 Skill、9 支 auto-edit Skill），最後探測生圖能力。當使用者說「幫我檢測環境」「檢測環境」「安裝環境」「裝工具」「Skills 看不到 auto-edit」時使用。已經在剪片、只是某個剪片指令出錯時不要用（看那個階段 Skill 的退路表）。 version: 0.1.0 --- # 環境準備 規則在 AGENTS.md；本檔只寫這一階段的指令、判準、停點、退路。 衝突時：使用者當下的話 > AGENTS.md > shared/style.md > USAGE.md > USAGE-reference.md > 本 Skill；派工時派工訊息 > 本 Skill。 | 輸入 | 指令 | 產出 | 完成判準 | 停點 | |---|---|---|---|---| | — | `bash setup/check.sh`（Mac）、`setup\check.ps1`（Windows） | 每一步的進度行、總表、能力表 | 檢測 JSON 的 `all_ok` 是 `true` | 沒有（安裝清單問一次同意） | 手冊是 `setup/README.md`（步驟代號 M＝`setup/mac.md`、W＝`setup/windows.md`、S＝共用步驟）；本檔不抄手冊，只指路、補完成判準。進度回報格式照 `setup/README.md`「進度回報格式」原文。 ## 開始前 | 條件（看得到的事實） | 做法 | |---|---| | 這一輪還沒讀過 `setup/README.md` | 先讀全文 | | 使用者只說「Skills 看不到 auto-edit」 | 只做 S4（`node setup/install-skills.mjs`），再跑一次檢測 | | 檢測顯示 `restart` | 照 `setup/README.md` 步驟 2：先加 PATH 前綴把 S1–S4 做完（前綴照手冊步驟 3 那一行），再請使用者重開 MiniMax Code（Mac `M8`、Windows `W9`），這一輪到這裡結束。重開後還是 `restart`，就整輪加前綴做，總表最後註明「之後的指令要加 PATH 前綴」 | | 檢測 JSON 的 `all_ok` 是 `true` | 不用列清單、不用問同意，直接從步驟 3 的 S5 做 | ## 步驟 **0. 檢測** ``` Mac：bash setup/check.sh Windows（PowerShell）：[Console]::OutputEncoding = [System.Text.Encoding]::UTF8; powershell -NoProfile -ExecutionPolicy Bypass -File setup\check.ps1 ``` 完成判準：最後一行 `JSON:` 解析成功，讀到 `todo` 與每一項的 `status`、`step`。 常見錯誤：Windows 的 shell 是 Git Bash → 改用 `powershell.exe -NoProfile -ExecutionPolicy Bypass -File setup/check.ps1`。 **1. 列清單、問一次同意**（AGENTS.md 規則 3：整份清單問一次） ``` （不用指令：照 setup/README.md 步驟 1 的範例，每步寫用途和指令） ``` 完成判準：回覆最後一行是 `❓ 請回：「可以開始」或要拿掉的步驟編號`。 常見錯誤：一項一項問 → 整份清單只問一次。