t0mtaylor/peepshow — reverse-engineered prompt

Reverse engineered prompt

GitHub

Build me a Node command line tool called peepshow that lets an AI inspect videos and animated images. I want to give it an mp4, mov, gif, APNG, or animated WebP, and have it pull out the important still frames as a timeline so I can paste or pipe the result into an LLM.

It should also extract audio when the file has it, make a small audio file, and optionally transcribe it using local Whisper if available, or a cloud provider if the user chooses one. The output should work as a readable summary by default, with options for JSON and a compact LLM friendly format.

Please include a Claude Code style workflow where dropping or mentioning a video path can trigger the frame extraction automatically, plus a normal CLI command for everyone else. Add a simple sink system so the full payload, video info, tags, frames, audio, and transcript can be sent to storage or other tools. Look up current docs online if needed.

Want more depth? Deep Reverse