Astra: research 15 YouTube tutorials then apply and draft SKILL.md

I want to complete this task: [SPECIFIC TASK].

My materials and requirements: [FILES, LINKS, OR DESCRIPTION].

Find practical YouTube tutorials relevant to this task. Prioritize videos from the past 6 months with high view counts and engagement.

Identify reuploads, retellings of the same source, and substantially duplicated content. After removing duplicates, keep up to 15 videos. Don't add irrelevant videos just to reach that number.

Use yt-dlp to download subtitles without downloading the videos. If it isn't installed, install it for the current operating system. Tell me if you hit a permissions barrier. Prefer manually created subtitles, falling back to auto-generated ones. If retrieval fails, try the video page's "Show transcript" panel. If you still can't get the transcript, say so and skip that video. Don't fill in content based only on its title.

Extract actionable steps, common mistakes, and verification criteria from the transcripts. Keep the source links.

Merge repeated methods, favoring the most complete explanations with concrete steps and conditions for when to use them. Multiple retellings of the same source do not count as independent evidence. Popularity does not make a method correct. When recommendations conflict, explain when each applies.

Apply the relevant methods to my materials and complete the task. Check the result against the verification criteria. State which checks passed and which you couldn't verify.

After I've reviewed the result and finished my edits, turn the useful steps, fixes, and verification criteria into a SKILL.md draft. Keep the sources and distinguish methods we actually verified from recommendations that remain untested. Show me the draft first. Save it only after I approve it, so we can reuse it for similar tasks.