3-hour podcasts/videos, highlights in 15 minutes.

Created a skill for Claude Code, just drop in a link — supports Xiaoyuzhou, YouTube, Bilibili, and works in both Chinese and English.
Each of the three AI models handles its own task:
• Claude manages the entire workflow
• Whisper converts audio to text
• Gemini processes long texts of up to 50,000 words and outputs structured summaries
The most interesting part is that the methods for obtaining audio on the three platforms are completely different. Xiaoyuzhou is the simplest, with audio links directly hidden in the page source code. YouTube has anti-scraping mechanisms, so it requires some workarounds. Bilibili is the most troublesome — standard methods are all blocked, so I had to directly call its underlying API to get the audio.
In testing, three videos (see the picture), 117min + 181min + 114min, all worked smoothly. The longest one, 181 minutes, transcribed over 50,000 words.
Previously, a 3-hour podcast could only be listened to or skipped. Now there's a third option: watch the highlights first, and if it’s worth it, go back and listen to the full original.