youtube-transcript-note

name	youtube-transcript-note
description	Create a structured Obsidian vault note from a YouTube video transcript — fetches captions with timestamps, cleans the text, divides into sections, adds a TOC and per-section video links, uploads the thumbnail to Imgur, and commits. Use when the user shares a YouTube URL and wants a full transcript note in the vault.

YouTube Transcript Note Workflow

Given a YouTube URL, run the following steps in order. Carry outputs forward — do not prompt the user between steps unless a step fails.

Step 1: Fetch the timestamped transcript

Run benevolentbadger with SRT format to get captions with timestamps:

uv run --no-active --project /Users/mtm/pdev/taylormonacelli/benevolentbadger benevolentbadger --format srt "<youtube_url>"

Save the full SRT output — you will need the timestamps in step 5.

Also run with plain text to get the raw transcript for cleaning:

uv run --no-active --project /Users/mtm/pdev/taylormonacelli/benevolentbadger benevolentbadger "<youtube_url>"

Step 2: Clean the transcript

This is a transcription task, NOT a summarization task. Every piece of spoken content must appear in the output — do not skip, condense, or summarize any portion of the transcript. If the transcript is long, output all of it.

Apply these rules to the plain-text transcript:

• Correct all spelling errors
• Fix grammatical mistakes (subject-verb agreement, tense consistency, etc.)
• Add proper punctuation (periods, commas, apostrophes, contractions)
• Capitalize proper nouns and sentence beginnings
• Place each complete sentence on its own line
• Separate each sentence with a blank line
• Combine fragmented phrases into complete, coherent sentences
• Remove filler words (um, uh, like, you know, so, kinda, gonna, wanna — replace contractions like "gonna" with "going to")
• Remove false starts and repeated words
• Remove narrator labels like [Narrator], stage directions like (laughing), and any trailing URLs

Step 3: Divide into sections

Read the cleaned transcript and identify 4–7 natural topic breaks.

Every sentence from the cleaned transcript must appear under exactly one section — nothing may be dropped or omitted.

For each section:

• Write a ## heading that names the topic (title case, concise)
• Place ALL cleaned sentences for that topic under the heading

Using the SRT output from step 1, find the start time (in seconds) for the first caption that falls under each section. Add the video timestamp URL as a bare link on the line immediately after the ## heading:

## Section name

https://youtu.be/<VIDEO_ID>?t=<seconds>s

First sentence of section.

Step 4: Build the TOC

At the top of the body (before the first ## section), add a table of contents using Obsidian wikilink heading format — not standard markdown anchors:

## Table of contents

- [[#Section one]]
- [[#Section two]]
- [[#Section three]]

Step 5: Upload hero image to Imgur

Invoke the social-to-imgur skill using the YouTube URL to get a permanent Imgur URL for the thumbnail.

Collect from the output:

• The permanent Imgur URL (e.g. https://i.imgur.com/xxxxxxx.jpeg)

Step 6: Derive the vault filename

Convert the video title to a vault-safe filename:

• All lowercase
• Words separated by spaces (no hyphens or underscores)
• Strip special characters (pipes, colons, slashes, etc.)

Prefix with the channel/creator name when it helps distinguish the note. Example: tasty kitchen cast iron skillet care and cooking.md

Step 7: Write the vault note

Write the note to /Users/mtm/Documents/Obsidian Vault/<filename>.md with this structure:

---
tags:
  - <relevant-tag>
pic: <imgur_url>
---

![<filename without .md>](<imgur_url>)

https://youtu.be/<VIDEO_ID>

<intro sentences — any content before the first natural section break>

## Table of contents

- [[#Section one]]
...

## Section one

https://youtu.be/<VIDEO_ID>?t=<seconds>s

<sentences>

## Section two

https://youtu.be/<VIDEO_ID>?t=<seconds>s

<sentences>

Rules:

• No # Title heading — the filename is the title in Obsidian
• Tags: use only pre-existing vault tags; do not mint new ones without user approval
• One sentence per line, blank line between sentences throughout
• No italics or bold anywhere in the file
• Link named products to their local vault page (e.g. [[flaxseed oil]]) if one exists, or to a Google Image Search URL otherwise

Step 8: Commit

Stage and commit the new file:

git add "<filename>.md"
git commit -m "Add <short topic> note from YouTube transcript"

Run each git command as a separate Bash call (no chaining with &&).

Step 9: Apply recipe cleanup (cooking/recipe content only)

Assess whether the video content is recipe or cooking-focused by examining any of the following signals:

• The video title contains words like: recipe, sauce, how to make, cooking, bake, roast, stir-fry, marinade, dipping, dumpling, noodle, soup, fry, etc.
• The channel name is associated with food or cooking
• The transcript contains ingredient measurements, quantities, or cooking technique vocabulary

If any signal indicates recipe/cooking content, invoke the recipe-cleanup skill, passing the path to the vault note just written:

/recipe-cleanup <absolute path to vault note>

The skill will fetch the printable recipe from the YouTube description (if one exists), add structured ### Ingredients checklists and ### Instructions blocks, update frontmatter with a creator field, and resolve ingredient wikilinks.

After the skill completes, commit the updated file:

git add "<filename>.md"
git commit -m "Apply recipe cleanup to <short topic> note"

If the content is not recipe or cooking-focused, skip this step entirely — do not invoke recipe-cleanup.

Final report

After all steps complete, report:

• The vault note filename
• The Imgur thumbnail URL
• The section breakdown with their video timestamps
• Whether recipe cleanup was applied