Self-Learn

Autonomously master a book through a 3-phase pipeline: extract atomic concepts, validate against NotebookLM, and extend into new domains.

/self-learn runs a fully autonomous 3-phase pipeline that reads an epub, extracts atomic knowledge into your Obsidian vault, validates that knowledge against NotebookLM until 100% mastery, then extends it into synthesis and applied research. All knowledge lives in Obsidian — never inside the skill itself.

Usage

# Full autonomous run
/self-learn --book <epub_path> --notebook-id <nlm_id>

# Run a specific phase only
/self-learn --phase extract --book <epub_path>
/self-learn --phase validate --notebook-id <nlm_id>
/self-learn --phase extend

How It Works

1. Phase 1: Extract

   /self-learn parses the epub chapter by chapter. For each chapter it identifies atomic concepts — named frameworks, core principles, step-by-step processes, mental models, key distinctions, and actionable tools — then writes one Obsidian note per concept (~100 words).

   Concepts are organized into topic folders and linked to related concepts from earlier chapters. After all chapters are processed, the skill reports the total number of topics extracted and their category breakdown.

2. Phase 2: Validate

   This phase runs an autoresearch loop using 4 independent roles with strict information barriers:

   Role	Responsibility
   Question Generator	Creates questions without seeing any answers
   Knowledge Agent	Answers only from your Obsidian notes — never from the book directly
   Oracle (NotebookLM)	Answers from the full book via notebooklm ask
   Judge	Scores answer similarity on a 0–100 scale without knowing which is agent vs oracle

   The loop generates max(100, num_topics × 3) questions — 50% topic-based, 30% blind (from chapter titles), and 20% adversarial. Any question scoring below 95 triggers a targeted update to the relevant Obsidian notes. The loop repeats with fresh questions until 100% of questions pass at ≥95.

   Wrong concepts are treated as catastrophic. The skill prioritizes fixing conceptual errors before filling missing details — a note that says the wrong thing is far worse than a note that’s incomplete.

   Example validation log:

   Round 1: 85/100 passed (85%) — weak: [business-operations, leadership-hats]
   Round 2: 94/100 passed (94%) — weak: [accountability-structures]
   Round 3: 100/100 passed (100%) — MASTERY ACHIEVED
   
3. Phase 3: Extend

   Once mastery is achieved, /self-learn extends the knowledge beyond the source book:

   • Synthesis — connects concepts to adjacent thinkers; notes tagged source: "synthesis"
   • Applied domains — applies ideas to new contexts; saved in applied/{domain}/, tagged source: "application"
   • Research — captures current developments related to the book’s themes; tagged source: "research-{date}"

   Extension notes use the same note template as Phase 1 but with different source tags.

After all three phases complete, the pipeline continues autonomously into /ingest → /audit → /absorb → /sync.

Knowledge Output Location

Extracted notes are written to your vault under the book’s slug:

{vault}/knowledge/raw/{book-slug}/
  ├── decision-making/
  ├── business-operations/
  ├── leadership/
  └── [discovered categories]/

Note Template

Each extracted concept is stored as a standalone Obsidian note:

---
source: "{book_title}"
author: "{author}"
chapter: {chapter_number}
tags: [{tags}]
---

# {Concept Title}

{~100 words: concise, crystal clear, key point only}

## Key Insight
{One sentence takeaway}

## Related
- [[related-concept-1]]
- [[related-concept-2]]

NotebookLM Setup

Before running the validate phase, you need to point the skill at your NotebookLM notebook. Run these two commands to set the active notebook and verify it responds:

~/.local/bin/notebooklm use {notebook_id}
~/.local/bin/notebooklm ask "{question}"

Replace {notebook_id} with your NotebookLM notebook ID (passed via --notebook-id).

Checking Status

Always use the summary.py script to check phase status — never generate the summary manually:

# Full status overview
python3 ${CLAUDE_PLUGIN_ROOT}/skills/self-learn/scripts/summary.py {vault}/knowledge/raw/[book-slug]

# Phase status only
python3 ${CLAUDE_PLUGIN_ROOT}/skills/self-learn/scripts/summary.py {vault}/knowledge/raw/[book-slug] --status

# Phase 3 details
python3 ${CLAUDE_PLUGIN_ROOT}/skills/self-learn/scripts/summary.py {vault}/knowledge/raw/[book-slug] --phase3

/self-learn requires Python 3.12+ and the notebooklm-py package. See the dependencies page for installation instructions.

Use /chain to run the entire pipeline — self-learn through sync — in a single command without manually triggering each phase.