Skip to content

v0.1.0 — Video Reverse Engineering

Latest
Latest

Choose a tag to compare

View all tags
@whaojie797-design whaojie797-design released this 05 Aug 05:53
· 1 commit to main since this release
v0.1.0
0d57c55

What it does

Agent Skill that reverse-engineers how a video was made. Given a local video (or a URL that yt-dlp can fetch), it:

  1. Extracts real keyframes (scene-cut detection, not blind sampling), subtitles, and audio track metadata.
  2. Produces a shot list with timestamps and what each shot is doing.
  3. Generates ready-to-paste AI prompts — image-generation prompts per keyframe plus a video-generation prompt that captures motion/camera/style.
  4. Writes a replication guide covering both AI-generated and live-action production paths.

Highlights

  • Cross-platform dependency install (macOS / Linux / Windows) — ffmpeg via winget/brew/apt, yt-dlp via pip.
  • ffmpeg version-adaptive frame extraction (-fps_mode vfr on new builds, -vsync vfr on older ones).
  • Host-agnostic: works under Codex, Claude Code, and Cursor — no hardcoded tool or product names.
  • Ships with a validator (tools/validate_skill.py) and a GitHub Actions smoke test that builds a synthetic 3-cut video and asserts keyframe extraction works end-to-end.

Install

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

See the README for the full workflow, before/after examples, and limitations.

Assets 2
Loading