whaojie797-design/video-reverse-engineering

Agent Skill: reverse-engineer how a video was made. Extracts real keyframes/subtitles/audio, then produces a shot list, ready-to-paste AI image+video prompts, and a full replication guide (AI and live-action paths).

Project Overview项目介绍

This agent skill enables AI coding agents to reverse-engineer how a video is created, supporting three popular agent platforms: Codex, Claude Code, and Cursor, and it is also compatible with DSH. To install the skill on your preferred agent, you just clone the repository to the agent’s skills directory using a simple one-line bash command, with a different command for each supported platform. After installation is complete, the agent will automatically detect the skill and read its configuration without any extra registration or manual setup steps.

Once installed, the skill triggers automatically when you send a video URL and ask how the video was made, so there is no need to manually call the skill by name. It is purpose-built for content creators and video makers who want to replicate the style, structure, and production flow of an existing video, rather than just getting a generic content summary of what the video is about. After extracting the video data, the skill analyzes each shot using a standardized set of terminology, then outputs ready-to-use assets that work for both AI generation and real-world on-location shooting.

The skill requires two external dependencies, ffmpeg for media processing and yt-dlp for video downloading, and it will automatically check for these tools on first run and prompt you with install instructions for your operating system if any are missing. It has several known limitations: it cannot download videos behind login walls or region locks, scene detection may miss slow camera movements, and it requires the host agent to have multimodal image reading capabilities to work correctly. The project is released under the permissive MIT open source license, so you can use, modify, and distribute it freely with no commercial restrictions.

这是一款面向AI Agent的视频逆向拆解与复刻技能,支持Codex、Claude Code、Cursor,同时适配DSH。传入一个视频链接后,它会先通过场景检测算法提取真实关键帧,获取精确到秒的时间码、字幕和音频信息,再输出完整分镜脚本、可直接复制使用的AI生图生视频提示词,以及包含AI生成和实拍两条路径的完整复刻指南,解决了普通Agent不看画面就编造镜头细节的问题。

它面向需要拆解视频创作逻辑、复刻视频内容的创作者,安装方式根据使用的Agent宿主不同,只需要把仓库克隆到对应的skills目录即可,无需额外注册,宿主会自动扫描读取配置。安装完成后,用户只需要发送视频链接并询问视频的制作方式,就能触发该技能,无需手动调用技能名称,还可以手动运行抽帧脚本调整检测参数适配不同类型的视频。

该技能依赖ffmpeg和yt-dlp两个外部工具,运行前需要先安装对应依赖,启动时会自动检查缺失工具并提示安装方法。它存在一些局限性,比如遇到平台反爬、登录限制时下载会失败,慢速运镜可能漏检镜头,复杂运镜判断可能出错,且依赖宿主的多模态读图能力。该项目采用MIT许可开源,可免费使用修改。

Pre-install check安装前体检Compatibility · Security兼容性 · 安全性 1 warning1 项注意
  • Only 2 stars - very few users, little community feedback星标只有 2,几乎没人在用,遇到问题缺少社区反馈
  • Not DSH-native: a multi-platform tool that may require Node / Electron or another runtime first非 DSH 原生,是多平台兼容工具:可能要先装 Node / Electron 等运行时
DSH walks through these 9 checksDSH 会逐条核对这 9 项

Compatibility兼容性

  • DSH, Node, OS and profile requirementsDSH 版本 / Node 版本 / 操作系统 / profile 是否满足要求
  • External dependencies and runtimes (Electron / Python / Docker, ...)外部依赖与运行时(Electron / Python / Docker 等)是否齐备
  • Conflicts with installed plugins: command names, skill / tool names, ports, duplicate MCP registration与已装插件是否冲突:命令名、skill / tool 重名、端口占用、重复 MCP 注册

Security安全性

  • Repo matches the facts registered here; archived or abandoned?仓库是否与页面登记一致,是否归档或长期停更
  • Safety of preinstall / install / postinstall and install.sh / setup.ps1preinstall / install / postinstall 与 install.sh、setup.ps1 是否安全
  • curl|bash, download-then-execute, obfuscation, unrelated domains → stop immediatelycurl|bash、下载即执行、混淆代码、无关域名 → 立刻停止
  • Typosquatting or unmaintained packages among the new dependencies新增依赖里有没有 typosquatting 或无人维护的包
  • Requested permissions vs. what the feature actually needs申请了哪些权限、是否超出功能所需(filesystem / network / shell / clipboard)
  • Any sudo / admin requirement, plus uninstall and rollback是否要求 sudo / 管理员权限,以及卸载与回滚方式

Anything uncertain must be marked unknown with a note on how to confirm it. This site's signal screen is a static snapshot, not a security audit.拿不准的必须标「未知」并说明要我怎么确认。本站的信号筛查是静态快照,不能替代安全审计。

Or use CLI install (for developers)或使用命令行安装(适合开发者)

CLI Install命令行安装

dsh plugin --profile web add github:whaojie797-design/video-reverse-engineering

把 whaojie797-design/video-reverse-engineering 加入你的 DSH 配置(web profile)即可启用。

READMEREADME

video-reverse-engineering

validate License: MIT version

视频逆向拆解与复刻 —— 给 AI Agent 用的 Skill。丢一个视频链接进去,得到分镜脚本、可直接用的 AI 生图/生视频提示词、以及一份完整复刻指南。

Reverse-engineer how a video was actually made. Give an agent a video URL; get back a shot list, ready-to-paste AI image/video prompts, and a full replication guide — grounded in real extracted frames, not in the model's imagination.


为什么需要它

让 Agent "分析一下这个视频",十次有九次拿回来的是内容摘要——它讲了什么、观点是什么。但如果你的目的是照着拍一条,内容摘要一点用都没有。

更糟的是,大多数 Agent 根本没看过画面,就开始输出"镜头 1:一个咖啡店的空镜,氛围温馨"这种谁都写得出来的废话。没有时间码、没有景别、没有运镜、没有色调,拿去生图生不出东西,拿去实拍也不知道怎么拍。

Before(直接问 Agent) After(装上这个 Skill)
分析依据 凭标题和简介脑补 场景检测算法抽出的真实关键帧,逐张看过
时间码 没有,或者是估的 从 timestamps.txt 取,精确到秒
镜头描述 "一个温馨的空镜" 景别/角度/运镜/构图/光线/调色/转场,术语表统一用词
能不能直接用 不能 制图 Prompt 可直接粘进 Midjourney / 即梦 / 可灵
复刻路径 没有 AI 生成 + 实拍两条路径都给,含后期与平台适配
编造风险 高 明确禁止:没看到画面就不许写镜头内容

它怎么工作

视频 URL
   ↓  scripts/extract_shots.sh
下载 → 场景检测抽帧 → 抽字幕 → 抽音频 → 读元数据
   ↓
frames/shot_0001.jpg ... + timestamps.txt(精确时间码)
   ↓  宿主逐张读图,对照 references/shot_glossary.md 的术语
逐镜头视觉分析(景别/角度/运镜/构图/光线/调色/主体/文字/转场)
   ↓  + 节奏结构与音频分析
   ↓  对照 references/output_templates.md
┌──────────────┬──────────────────┬──────────────┐
│ A. 分镜脚本   │ B. 制图提示词      │ C. 复刻指南   │
│ 逐镜头表格    │ 中文说明+英文Prompt │ AI路径+实拍路径│
└──────────────┴──────────────────┴──────────────┘

关键设计:抽帧和逐帧分析这两步不能跳过。即使用户只要分镜脚本,也必须先真的把帧抽出来看过——这是所有产出的地基。素材抓不到时有三级降级方案(要用户发文件 → 找现成拆解文章辅助 → 只用标题简介并如实告知分析会打折),但底线是绝不在没看到画面的情况下编造镜头。


安装

三个宿主分别装,选你在用的那个:

Codex

git clone https://github.com/whaojie797-design/video-reverse-engineering.git ~/.codex/skills/video-reverse-engineering

Claude Code

git clone https://github.com/whaojie797-design/video-reverse-engineering.git ~/.claude/skills/video-reverse-engineering

Showing the opening section of the README — the full document lives in the repository以上为 README 开头摘要,完整文档在仓库内 · View the full README on GitHub →在 GitHub 查看完整 README →

← 上一个 Prev dsh-download-progress 下一个 Next obsidian-workbench →