bradautomates/claude-video ? reverse-engineered prompt

I want a way for Claude to actually watch and analyze any video I give it—whether it’s a YouTube link, a screen recording, or a file on my computer. When I ask a question about a video, it should automatically download it, pull out the important frames and the transcript (using captions if available, or Whisper AI if not), and give Claude all that info so it can answer as if it really watched and listened. I want to be able to ask things like “what’s happening at the 30 second mark” or “summarize this video,” and get real answers grounded in the audio and visuals, not just guesses from the title or sparse transcripts. Make it easy to set up—install the required tools automatically and explain if I need to add an API key for the Whisper fallback. Feel free to look up any docs you need online.