seanlu2006/vid-for-agents ? reverse-engineered prompt

Build me a simple macOS command line tool that turns a video into something an AI can actually read.

I want to run one command on either a local video file or a YouTube or Instagram link, and get back a new folder with sampled still frames, a full transcript, and a small README that tells an agent what to look at first. It should work completely on my machine, so nothing gets uploaded and there is no API cost.

If the link needs cookies, support that too. If the video has no audio, still extract the frames and explain that the transcript is missing. If it is audio only, just give me the transcript. Include the original metadata when the source is a URL, and make the output easy for Claude or Cursor to read without extra setup.

Use current local tools if needed, and keep the install as painless as possible for a normal Mac user.