course

Process mode (default): Crawl an online course page, extract the lecture schedule with all materials (slides PDFs, YouTube videos, readings), then process each lecture into a detailed vault-formatted note with embedded slide images and synthesized transcripts. Creates a Course index note linking everything together.

Refine mode: Re-read existing course lecture notes and improve them — deepen equation explanations, add missing analogies, fix structure, strengthen cross-references. Triggered when user says "refine", "improve", "fix", or references existing course notes rather than a URL.

This is an orchestrator — it coordinates downloading, extraction, and synthesis across multiple lectures, producing a complete course knowledge package in the vault.

<Use_When>

• User shares a course URL (syllabus, schedule, or homepage)
• User says "process this course" or "take notes on these lectures"
• User pastes a university course page with a lecture list
• User wants to batch-process a series of lectures from one course
• User has a YouTube playlist of course lectures
• User wants to refine/improve existing course notes ("refine mit diffusion notes", "improve L03") </Use_When>

<Do_Not_Use_When>

• User has a single YouTube video (use /youtube)
• User has a single local video file (use /lecture)
• User wants to process an existing vault note that's NOT a lecture (use /process)
• User wants a single paper summarized (use /paper) </Do_Not_Use_When>

Stage 1: CRAWL — Extract the Lecture List

Parse the course URL from $ARGUMENTS. If no URL, ask the user.

Fetch the course page and extract structured lecture data:

WebFetch(
  url="COURSE_URL",
  prompt="Extract the complete course structure as JSON. For each lecture/session, include:
    - number (int)
    - title (string)
    - date (string, if available)
    - slides_url (string or null — look for PDF links to slides/lecture notes)
    - video_url (string or null — look for YouTube links)
    - readings (array of {title, url} — papers, blog posts, textbook chapters)
    - description (string or null — any summary text)
  Also extract:
    - course_title (string)
    - course_code (string or null)
    - instructors (array of strings)
    - course_url (string — the page URL)
    - course_notes_url (string or null — if there's a single PDF of all course notes)
  Return ONLY valid JSON, no markdown fencing."
)

Parse the JSON response. If the page has relative URLs for slides/videos, resolve them against the course URL's base.

Handle edge cases:

• If the page is a YouTube playlist: extract video IDs and titles from the playlist
• If slides URLs are relative (e.g., ../docs/lecture_01.pdf): resolve to absolute URLs
• If no structured schedule found: ask user to provide lecture list manually