Barty-Bart/motion-graphics ? reverse-engineered prompt

Build me a Claude Code skill that takes a video and a transcript, then creates motion graphics B roll that lines up with the spoken words.

I want it to inspect the video, figure out where the speaker and any picture in picture areas are, estimate timing from the transcript, and then plan a few clips for me to approve before rendering. Each clip should feel like one continuous animated shape that keeps transforming instead of cutting around, like a pill turning into a card, a terminal, or a chart. Some clips should be full frame cutaways, and some should be transparent panels that sit beside the speaker.

After it renders, I want a preview of the video with the graphics added, a compare page so I can review original versus edited, a simple viewer for stepping through each clip, and a timing file that shows what each clip covers. Keep it working with current docs if you need to look things up, and make sure the output is something I can drop into my normal editing workflow.