speechlab0210/video-production-skill ? reverse-engineered prompt

Build me a reusable skill for an AI coding agent that can make narrated educational videos almost end to end. I want the workflow to take a topic, write a script, make slide visuals in either a hand drawn whiteboard style or simple HTML slides, generate voiceover, check each spoken line with speech recognition so the audio really matches the script, assemble the final video, add well timed subtitles, and export a cover image too.

Please make it practical, with one main instruction file that the agent must read first and follow exactly, plus runnable scripts for each stage, a small demo example, and notes about common mistakes from real production use. Aim it at Traditional Chinese first, especially pronunciation issues and subtitle timing, but keep the overall process usable for other languages too. Assume I can provide the needed API keys, a voice ID, and local video tools. If anything is unclear, look up current docs online, but keep the setup simple and reliable.