# Extract Workflow

Extract dynamic, content-adaptive wisdom from any content source.

## Input Sources

| Source | Method |
|--------|--------|
| YouTube URL | `fabric -y "URL"` to get transcript |
| Article URL | WebFetch to get content |
| File path | Read the file directly |
| Pasted text | Use directly |

## Execution Steps

### Step 1: Get the Content

Obtain the full text/transcript. For YouTube, use `fabric -y "URL"` to extract transcript. Save to a working file if large.

### Step 2: Deep Read

Read the entire content. Don't extract yet. Notice:
- What domains of wisdom are present?
- What made you stop and think?
- What's genuinely novel vs. commonly known?
- What would {PRINCIPAL.NAME} highlight if he were reading this?
- What quotes land perfectly?

### Step 3: Select Dynamic Sections

Based on your deep read, pick 5-12 section names. Rules:
- Section names must be conversational, not academic
- Each must have at least 3 quality bullets
- Always include "Quotes That Hit Different" if source has quotable moments
- Always include "First-Time Revelations" if genuinely new ideas exist
- Be SPECIFIC — "Agentic Engineering Philosophy" not "Technology Insights"

### Step 4: Extract Per Section

For each section, extract 3-15 bullets. Apply tone rules from SKILL.md:
- 8-20 words, flexible for clarity
- Specific details, not vague summaries
- Speaker's words when they're good
- No hedging language
- Every bullet worth telling someone about

### Step 5: Add Closing Sections

Always append:
1. **One-Sentence Takeaway** (15-20 words)
2. **If You Only Have 2 Minutes** (5-7 essential points)
3. **References & Rabbit Holes** (people, projects, books, tools mentioned)

### Step 6: Quality Check

Run the quality checklist from SKILL.md before delivering.

### Step 7: Output

Present the complete extraction in the format specified in SKILL.md.