waybout107-cyber/video-recap-skills ? reverse-engineered prompt

Build me this project so I can point it at one or more video files and get back a Chinese narration recap video.

I want the flow to feel simple, like I give it a source video and a target length, then it handles scene detection, speech recognition, understanding what is happening on screen, writing the recap script, generating the voiceover, mixing everything together, and exporting a finished video with subtitles. If it can also produce an editable draft for JianYing or CapCut, that’d be great.

Please make sure it works locally with ffmpeg and a single MiMo API key, without needing a GPU. Add clear setup and usage steps, a small example I can run, and helpful checks if ffmpeg or the API key is missing. If you need to look up current docs online for anything, go ahead.