zenstory-ai /video-recap-skills

About video-recap-skills

Turn one or more videos into a finished Chinese narration recap (视频解说) with six agent skills. Run it inside a coding agent like Claude Code or Codex CLI. Local ffmpeg and a paid Xiaomi MiMo API key are all it needs. It handles scene detection, speech recognition, editing, narration, voiceover and mixing, then exports an editable JianYing/CapCut draft — no GPU, no model downloads.

Best for: Recap channels that want an agent pipeline turning long videos into narrated Chinese cuts.

• Chains six skills: understanding, script, cut, voiceover, assemble, plus an orchestrator
• Fuses scene detection, frame extraction, ASR and VLM analysis into one timeline
• Decides per-beat audio ownership before writing any narration
• Cuts the source first, then writes narration against that finished timeline
• Exports timeline.json as an editable JianYing/CapCut draft
• Writes machine-readable QC files: narration lint, assembly QC, delivery QC

1. 1
   Install Python 3.10+, ffmpeg with libass on PATH, and get a Xiaomi MiMo API key
2. 2
   Set the key: the command below (add the next command for tp-* Token Plan keys)

   export MIMO_API_KEY=your-mimo-key

   Copy

   export MIMO_TOKEN_PLAN_CLUSTER=cn

   Copy
3. 3
   In Claude Code run the command below then the next command

   /plugin marketplace add zenstory-ai/video-recap-skills

   Copy

   /plugin install video-recap-skills@video-recap

   Copy
4. 4
   For Codex CLI run the command below and the next command

   codex plugin marketplace add zenstory-ai/video-recap-skills

   Copy

   codex plugin add video-recap-skills@video-recap

   Copy
5. 5
   Ask the agent to check the environment (Python, ffmpeg/libass, MiMo config)
6. 6
   Give a video path and what you want, e.g. make a Chinese narration recap of /path/to/video.mp4 with burned-in subtitles

• Video RecapTurns a film or episode into a Chinese narration recap video. It checks your setup, then runs the watch, script, cut, voice and assemble steps.