video-recap-skills
AI 视频解说 Agent 技能包:自动切镜、看画面写解说词、配音混音合成成片,可导出剪映/CapCut 草稿,无需 GPU | Agent skills for narrated video recaps & CapCut drafts
What is video-recap-skills?
AI 视频解说 Agent 技能包:自动切镜、看画面写解说词、配音混音合成成片,可导出剪映/CapCut 草稿,无需 GPU | Agent skills for narrated video recaps & CapCut drafts
Overview of video-recap-skills
What it is
Video Recap Skills is a set of seven skills (six production + one on-demand reference) designed for programming agents like Claude Code, Codex CLI, OpenCode, or OpenClaw. It turns one or more video files into a Chinese narration recap (视频解说). The pipeline includes scene detection, ASR, VLM, script writing, TTS, ffmpeg assembly, and optional editable JianYing / CapCut draft export. It requires only local ffmpeg and a remote MiMo API key; no GPU, no pip install, and no model downloads. Supported input formats: .mp4, .mov, .mkv, .webm.
Key features
- One key, local ffmpeg only. ASR, VLM, and TTS all go through Xiaomi MiMo; local side uses only Python standard library and
ffmpeg. TTS can be switched to Fish Audio. - Creative decisions first, voice assignment second. The agent compares feasible editing directions during initial planning, writes a story plan (
recap_story_plan.json) with viewer promise, POV, dramatic question, and per-beat changes, then assigns audio ownership (narration, original dialogue, or silence) for each beat. - Edit first, narrate later – timeline alignment is natural. The editing mode first cuts the long video into a finished clip, then writes narration against that clip. Multiple source videos can be combined by
source_id. - Editable output beyond the final video. The multi-track
timeline.jsoncan be exported as a JianYing / CapCut draft, allowing further editing of original clips, narration, BGM, subtitles, and image overlays. - Every step leaves a verifiable record. Narration lint, assembly QC, and final QC produce machine-readable reports; revision logs track changes, frozen items, and verification results.
Use cases
Turn a segment or multiple segments of video into a Chinese narration recap. For example, selecting clips from a short drama series, editing, writing a script, dubbing, mixing, and packaging into a final video with optional editable draft export. Suitable for video commentary, content repurposing, and creative storytelling.
Getting started
Prerequisites: Python 3.10+, ffmpeg on PATH, and a Xiaomi MiMo API key. For subtitle burning, ffmpeg with libass is needed (on macOS use brew install ffmpeg-full). Set environment variables: bash export MIMO_API_KEY=your-mimo-key export MIMO_TOKEN_PLAN_CLUSTER=cn # only for tp-* Token Plan keys: cn | sgp | ams Install via Claude Code: text /plugin marketplace add zenstory-ai/video-recap-skills /plugin install video-recap-skills@video-recap Or via Codex CLI: bash codex plugin marketplace add zenstory-ai/video-recap-skills codex plugin add video-recap-skills@video-recap For OpenCode and OpenClaw, see the repository README for detailed instructions.
License and pricing
The project is open source under the MIT license. It is free to use. Note that the MiMo API service is pay-as-you-go (approximately 1.3 CNY per full video in testing); Fish Audio TTS also has its own pricing.
Based on the project's public README and repository metadata.