nlm-book-deep-reader

name

nlm-book-deep-reader

description

6-prompt deep reading pipeline for books and long documents via Google NotebookLM. Follows the Stanford NotebookLM Workflow — sequentially executes Core Argument Extractor, Assumption Auditor, Personal Relevance Filter, Steelman Challenger, Action Extractor, and Permanent Note Builder. Produces Bloom's-taxonomy-aligned analysis that extracts deeper insights than reading a book twice, in ~20 minutes. Use when the user asks to "deep read a book", "analyze this book", "book deep reader", "Stanford NotebookLM workflow", "6-prompt book analysis", "deep reading pipeline", "book analysis with NotebookLM", "nlm-book-deep-reader", "책 심층 분석", "책 딥리딩", "북 딥리더", "NotebookLM 북 분석", "6단계 프롬프트 분석", "책 분석 파이프라인", "영구 노트 만들기", "Zettelkasten 분석", or wants to extract structured insights from a book/document using NotebookLM. Do NOT use for ad-hoc studio_create on existing notebooks — use notebooklm-studio. Do NOT use for accelerated learning with quizzes and study artifacts — use nlm-deep-learn. Do NOT use for academic paper review with PM analysis — use paper-review. Do NOT use for slides-only from local docs — use nlm-slides. Do NOT use for notebook CRUD without the 6-prompt pipeline — use notebooklm.

NLM Book Deep Reader

6-prompt deep reading pipeline for books and long documents via Google NotebookLM.

Based on the Stanford Students' Secret NotebookLM Workflow — a sequence of 6 prompts aligned to Bloom's Cognitive Hierarchy (Understand → Analyze → Apply → Evaluate → Create) that extracts deeper insights than reading a book twice, in approximately 20 minutes.

Prerequisites

• notebooklm-py CLI connected and authenticated
• Source material: PDF file path, URL, or pasted text
• For Prompt 3 (Personal Relevance Filter): user's context (role, challenges, goals, decisions)

Pipeline Overview

Source Upload ──► P1: Core Argument ──► P2: Assumption Audit ──► P3: Personal Relevance
                                                                         │
P6: Permanent Notes ◄── P5: Action Extractor ◄── P4: Steelman Challenge ◄┘

Phase	Prompt	Bloom Level	Output
1	Core Argument Extractor	Understanding	Core thesis + 3-5 sub-arguments with evidence strength
2	Assumption Auditor	Analysis	Unstated assumptions + vulnerability assessment
3	Personal Relevance Filter	Application	Applicable vs. non-applicable ideas for user's context
4	Steelman & Challenger	Evaluation	Strengthened argument + strongest counter-argument + verdict
5	Action Extractor	Synthesis	5 actionable changes ranked by impact/friction
6	Permanent Note Builder	Creation	5 Zettelkasten-style permanent notes

Execution

Phase 0: Setup — Create Notebook & Ingest Source

notebooklm create "Deep Read: <Book Title>"

Add the source material. Supported source types:

notebooklm source add <notebook_id> --file "<absolute_path_to_pdf>"
notebooklm source add <notebook_id> --url "<url>"
notebooklm source add <notebook_id> --text "<pasted_content>" --title "<title>"

Wait for source processing to complete before proceeding.

If the user has additional context documents (reviews, interviews, related articles), add them as supplementary sources. The primary book/document should be ingested first.

Phase 1: Core Argument Extractor

Goal: Identify the single core argument (not topic) the author is making, plus sub-arguments.

notebooklm chat <notebook_id> """
Read the entire document and identify the single core argument the author is trying to
make. Not the topic — the argument. The specific claim the author is trying to persuade
the reader of. State it in 2 sentences or fewer.

Then identify 3-5 sub-arguments that support this core argument. For each sub-argument:
1. State the claim clearly
2. Identify the evidence the author provides
3. Rate the evidence strength on a scale from Anecdote → Case Study → Survey Data →
   Meta-Analysis → Empirical Proof

Format the output as:
## Core Argument
[2 sentences max]

## Sub-Arguments
### 1. [Sub-argument title]
- **Claim**: ...
- **Evidence**: ...
- **Evidence Strength**: [rating] — [brief justification]

[repeat for each sub-argument]
""")

Save response as a note:

note(notebook_id, action="create", title="P1: Core Argument Analysis", content=<response>)

Phase 2: Assumption Auditor

Goal: Surface unstated assumptions and assess vulnerability of the core argument.

notebooklm chat <notebook_id> """
Identify all significant assumptions the author makes but does not explicitly state or
defend. Consider assumptions about:
- Human nature and behavior
- How organizations work
- How change happens
- Market dynamics or social systems
- What the reader already believes

For each assumption:
1. State the assumption clearly
2. Assess: Is it supported by external evidence? Challenged by field experts?
   Or is it the author's worldview presented as universal truth?
3. Rate vulnerability: If this assumption is wrong, how much does it undermine the
   core argument? (Low / Medium / High / Critical)

Then answer: Which single assumption, if proven incorrect, would most undermine the
entire core argument? Why?

Format as:
## Identified Assumptions
### 1. [Assumption]
- **Category**: [human nature / organizational / change / market / reader belief]
- **Assessment**: [supported / challenged / worldview-as-truth]
- **Vulnerability**: [Low/Medium/High/Critical]
- **If wrong**: [impact on core argument]

[repeat for each]

## Most Critical Assumption
[Which one and why]

## Overall Robustness Assessment
[Does the book's argument survive assumption scrutiny?]
""")

Save response:

note(notebook_id, action="create", title="P2: Assumption Audit", content=<response>)

Phase 3: Personal Relevance Filter

Goal: Filter the book through the user's specific context to find directly applicable ideas.

Before running this prompt, ask the user for their context if not already provided:

• Current role and work
• Key challenges they face
• Goals they're pursuing
• Decisions they're facing

notebooklm chat <notebook_id> """
My specific context: <USER_CONTEXT>

Filter the entire document through my context above. Categorize every major idea,
framework, and recommendation into:

## Directly Applicable
Ideas and frameworks I can apply to my specific situation right now.
For each:
- **Idea**: [what it is]
- **How it applies**: [specific application to my context]
- **Implementation sketch**: [what it would look like in my real situation]

## Interesting but Not Applicable
Ideas that are intellectually valuable but don't map to my current situation.
For each:
- **Idea**: [what it is]
- **Why not applicable**: [specific reason it doesn't fit my context]
- **When it might become relevant**: [future scenario where it applies]

## Relevance Summary
- Total applicable ideas: [count]
- Highest-impact applicable idea: [which one and why]
- Surprising connection: [an unexpected way the book relates to my work]
""")

Save response:

note(notebook_id, action="create", title="P3: Personal Relevance Filter", content=<response>)

Phase 4: Steelman & Challenger

Goal: Stress-test the argument by making it stronger AND building the best counter-argument.

notebooklm chat <notebook_id> """
Perform a rigorous intellectual stress test of this book's core argument in three parts:

## Part 1: Steelman — Make the Argument Stronger
Take the author's core argument and make it STRONGER than the author did.
- Add the best supporting evidence the author didn't cite
- Identify the strongest version of this argument that could be made
- What real-world examples or data would make this argument even more compelling?

## Part 2: Steel-man Challenge — Build the Strongest Counter-Argument
Now construct the most powerful counter-argument possible:
- Who are the most credible opposing thinkers? What would they say?
- What real-world evidence contradicts the author's premise?
- What historical examples show the opposite conclusion?
- Where does the author's logic break down under edge cases?

## Part 3: Verdict Under Scrutiny
After performing both exercises:
- How well does the book's argument hold up under serious examination?
- What parts survive the challenge? What parts don't?
- Rate overall argument resilience: [Fragile / Moderate / Robust / Antifragile]
- What is the most important caveat a reader should keep in mind?
""")

Save response:

note(notebook_id, action="create", title="P4: Steelman & Challenge", content=<response>)

Phase 5: Action Extractor

Goal: Convert insights into specific, immediately actionable behavioral changes.

notebooklm chat <notebook_id> """
Based on the book's core ideas AND my personal context (<USER_CONTEXT_BRIEF>),
generate exactly 5 changes I can implement within the next 30 days.

Requirements — each change must be:
- A specific behavioral change, not a vague direction like "be more strategic"
- Have a clear TRIGGER (when/where it activates)
- Have a clear ACTION (exactly what to do)
- Have a clear MEASUREMENT (how I know it's working)

Format each as:
### Change [N]: [Title]
- **Trigger**: [specific situation or cue]
- **Action**: [exactly what to do]
- **Measurement**: [how to track success]
- **Impact**: [High/Medium/Low]
- **Friction**: [High/Medium/Low]

Then:
## Ranking by Impact ÷ Friction
[Ordered list from best to worst ratio]

## Start Tomorrow
Pick the #1 ranked change and describe:
- **What exactly to do tomorrow morning**
- **What the first week looks like**
- **First checkpoint at Day 7: what to evaluate**
""")

Save response:

note(notebook_id, action="create", title="P5: Action Plan", content=<response>)

Phase 6: Permanent Note Builder

Goal: Synthesize into 5 Zettelkasten-style permanent notes that remain useful independently.

notebooklm chat <notebook_id> """
Synthesize the entire book into exactly 5 permanent notes following the Zettelkasten method.

Each note must be:
1. **Core Idea in My Own Words** — Explain the idea as if to someone who hasn't read the
   book. No jargon, no references to "the author says." Pure concept.
2. **Cross-Domain Connection** — Connect this idea to a concept from a completely different
   field (science, philosophy, engineering, psychology, economics, art, etc.). Show how
   the same pattern appears elsewhere.
3. **Best Supporting Evidence** — One compelling piece of evidence that makes this idea
   credible. Specific data point, study, or real-world example.
4. **Open Question** — One important question this idea raises but doesn't fully answer.
   Something worth continued thinking.

Format each note as:
---
## Note [N]: [Concept Title]

### Idea
[2-3 sentences, own words, standalone]

### Connection
[Link to concept from another domain + why the connection matters]

### Evidence
[One specific, compelling piece of supporting evidence]

### Open Question
[One question worth continued investigation]
---

Design these notes so they are independently useful — someone who never reads this book
should still gain valuable insight from any single note.
""")

Save response:

note(notebook_id, action="create", title="P6: Permanent Notes", content=<response>)

Phase 7: Synthesis & Output

After all 6 prompts complete:

1. Compile a summary of all 6 phases into a single Korean markdown document
2. Save to disk at outputs/book-deep-reader/{date}/{slug}/analysis.md
3. Post to Slack #효정-할일 with a 3-message thread:
   • Message 1: Book title + core argument (2 sentences) + overall verdict
   • Message 2: Top 3 applicable ideas + #1 action for tomorrow
   • Message 3: 5 permanent note titles as a quick reference index

Optional outputs (if user requests):

• Upload to Notion via md-to-notion
• Generate NotebookLM studio artifacts (audio overview, slides) via notebooklm-studio
• Generate NLM slide decks via nlm-dual-slides

User Context Collection

For Phase 3, if the user hasn't provided context, ask with this template:

📝 개인 맥락을 입력해주세요 (Personal Relevance Filter용):

• 현재 역할/업무: (예: AI 플랫폼 PM, 스타트업 CTO...)
• 주요 도전과제: (예: 팀 스케일링, 제품 시장 적합성...)
• 현재 목표: (예: 시리즈 A 펀딩, MAU 10만...)
• 직면한 의사결정: (예: 빌드 vs 바이, 시장 피벗...)

Flags & Options

Flag	Default	Description
--skip-personal	false	Skip Phase 3 (no personal context needed)
--skip-actions	false	Skip Phase 5 (analysis-only mode)
--with-studio	false	Generate NLM audio overview after analysis
--with-slides	false	Generate dual-audience NLM slide decks
--with-notion	false	Publish analysis to Notion
--lang	ko	Output language: ko (Korean) or en (English)

Error Handling

• If source_add fails: check file path is absolute; retry once; if still fails, try alternative source_type
• If notebook_query returns thin response: append "Please be more detailed and specific. Reference actual content from the source document." and re-query
• If user context is missing for Phase 3: pause and collect before proceeding
• If notebooklm CLI is unavailable: fall back to direct LLM analysis (degraded mode without source grounding)

Design Principles

• Sequential, not parallel: Each prompt builds on cumulative context in the notebook
• Save every response as a note: The notebook becomes a persistent artifact
• Korean output by default: Summary, Slack posts, and compiled document in Korean
• Bloom's alignment preserved: The 6 prompts follow Understanding → Analysis → Application → Evaluation → Synthesis → Creation
• Zettelkasten fidelity: Phase 6 produces genuinely reusable, connection-rich permanent notes