yt_trans · community-agentic-os
A YouTube playlist of tarot card-meaning videos, pulled into a searchable transcript corpus, then mined for the repeatable moves the teacher uses every time.
A lady named Kate makes videos that explain what each tarot card means. We took 8 of her videos and turned all her talking into words you can read and search.
So now, instead of watching for hours, you can read it fast and the computer remembers her tricks. 🎉
Seed playlist: Kate, "the daily tarot girl." Channel UCMMLTMTE90iVFwKHe9PAlXw.
Every transcript ships as verbatim .txt plus a line-indexed, timestamped .jsonl
so any downstream claim can cite the exact second it came from.
Aces and Twos of the Rider-Waite-Smith minor arcana, plus the Three of Pentacles. Segment count is the number of timestamped, individually citable caption lines.
9r-gFvBYbxc), is held out as
final at 8/9. YouTube IP-throttled its caption endpoint and the block did not clear across three
retry passes over roughly 75 minutes: youtube-transcript-api returned IpBlocked
and yt-dlp captions returned HTTP 429 on all four player clients
(web_safari, tv_embedded, android, mweb). The other eight stand as the corpus.Across the eight transcripts the pipeline counted how often each analytical move shows up. A move present in two or more videos becomes a registered workflow. Four map onto tools the Community OS already had. One had no home, so it became a new tool.
| Move | Frequency | Routes to |
|---|---|---|
| Quotable merch hook | 8 / 8 | etsy_products (reinforced) |
| Symbol & archetype | 8 / 8 | tarot_map (reinforced) |
| Critical argument | 7 / 8 | essay_draft (reinforced) |
| Scene-by-scene structure | 7 / 8 | scene_map (reinforced) |
| Emotional turn | 7 / 8 | NEW TOOL |
Kate pivots a card's meaning on an emotional beat in 7 of 8 videos, and no
existing OS tool watched for it. The pipeline generated emotional_turn.py into
community-agentic-os/tools/_generated/ and registered it. The tool scans an episode's
evidence for emotional-turn signal words and cites the exact line of every hit, so it is
gate-clean by construction: it cannot make an uncited claim.
python3 os.py analyze --ep <EP>
.txt is the primary
deliverable. Summaries run only because they feed a named downstream stage (workflow extraction),
where structured JSON beats raw captions.