Day with AI: YouTube Scrape Skill and Second Mac Setup. A Calm Wednesday.

What I worked on today

Wednesday. A calmer day, two smaller projects. But both needed.

1. YouTube scrape skill — yt-dlp instead of paid APIs

I wanted a way to extract transcripts and metadata from YouTube without paying for an API. Solution? yt-dlp.

The result is a `/youtube-scrape ` skill — give it a video link, it downloads subtitles (manual and auto-generated), extracts metadata and uses OpenAI to summarize everything into structured markdown. Output goes to `workspace/04-dokumentace/guides/youtube/`.

Fallback for videos without subtitles: whisper. But most videos have at least auto-generated subs, so it rarely triggers.

2. Setup script for a second Mac

I have a second Mac and manual setup is painful. Wrote a setup script that installs everything needed — brew packages, node, configs, tokens. From an hour of manual clicking to 26 minutes one-shot.

Time: AI vs without AI

Task	Without AI	With AI

YouTube scrape skill (yt-dlp + OpenAI)	4h	15min
Setup script for second Mac	1h	26min
Total	5h	41min

What I learned

• yt-dlp is incredibly powerful. Free, no API keys, handles subtitles, metadata and downloads. Ideal for a knowledge base pipeline.
• Setup scripts pay off even with just 2 machines. One hour of work saves hours on every future setup.