trellis

name	trellis
description	Grow an Obsidian knowledge graph from YouTube. Bulk-download a channel or playlist's transcripts and weave them into linked Markdown notes (topic tags plus [[wikilinks]]) whose Obsidian graph view clusters by subject. Use when the user wants to turn YouTube content into an Obsidian vault, a knowledge graph, linked notes, or a "second brain."

Trellis

A trellis gives a climbing plant something to grow on. This skill gives YouTube transcripts a structure to grow into: an Obsidian vault whose graph view clusters videos by topic. A downloader gets the words; Trellis turns them into a connected garden of notes.

Instructions

Step 1: Bulk-download the transcripts

Use a transcript downloader to write one file per video into a folder. The companion tool is TubeScribe:

tubescribe "https://www.youtube.com/@3blue1brown" -f txt -o transcripts/ -n 40

-n caps the count; drop it for the whole channel. Playlists work the same way. If live fetching is blocked (datacenter IP), run on the user's own machine or use TubeScribe's browser extension.

Optional, for human-readable note titles instead of video ids, build a titles.json ({ "<video_id>": "<title>" }) with yt-dlp:

yt-dlp --flat-playlist --print "%(id)s\t%(title)s" "<url>"

Step 2: Scaffold the vault

Turn each transcript into a Markdown note with frontmatter:

python scripts/build_vault.py --input transcripts/ --output vault/
# with titles:
python scripts/build_vault.py --input transcripts/ --output vault/ --titles titles.json

Each note gets title, video_id, source, type: transcript, an empty tags: [], and a "Related:" line ready for links.

Step 3: Weave the graph (the part that matters)

Edges in an Obsidian graph come from tags and [[wikilinks]]. This step is what turns a folder into a knowledge graph. Read the notes and edit them:

1. Tag each note by topic - 1-3 tags from a small, consistent vocabulary, in frontmatter tags: and/or as #inline-tags near the top (e.g. #linear-algebra, #calculus, #neural-networks). A tight vocabulary makes videos cluster cleanly.
2. Link related videos - on the Related: line add [[Note Title]] wikilinks to notes that share concepts: a video that builds on another, a series in order, a topic referenced across series. These cross-links are what make the graph interesting instead of just star-shaped.
3. Leave the transcript body untouched.

Work in batches for large channels. Prefer a few accurate links per note over many weak ones.

Step 4: Open in Obsidian

Tell the user:

1. Obsidian, Open folder as vault, pick vault/.
2. Open Graph View (Cmd/Ctrl + G).
3. In graph settings, under Groups, color by tag (e.g. tag:#calculus) to get the clustered look in examples/example-graph.html.

Notes

• Obsidian only graphs .md files, which the script already produces.
• Use -f json when downloading to keep timestamps for deep-linking into a note.
• examples/example-graph.html is a live preview of the end result.