pantheon-org/notebooklm

NotebookLM Research Assistant

Query Google NotebookLM to get Gemini's source-grounded answers from your uploaded documents. Each question opens a fresh browser session, retrieves the answer exclusively from your knowledge base, and closes.

When to Use This Skill

Trigger when the user:

• Mentions NotebookLM explicitly
• Shares a NotebookLM URL (https://notebooklm.google.com/notebook/...)
• Asks to query their notebooks or personal documentation
• Wants to add documentation to the NotebookLM library
• Uses phrases like "ask my NotebookLM", "check my docs", "query my notebook"

Mindset

NotebookLM is a grounding layer, not a search engine. Three principles govern every interaction:

1. Ground every answer in sources — answers come exclusively from uploaded documents; never synthesise from general knowledge when querying a notebook.
2. Never guess notebook content — use Smart Add (query first) to discover what a notebook contains before labelling it.
3. Follow up until complete — one query is rarely enough; check every answer for gaps and issue follow-ups before synthesising for the user.

When Not to Use This Skill

• The user needs general web search or LLM knowledge — use standard Claude responses instead
• No NotebookLM URL or notebook ID is available and authentication is not yet set up
• The user wants data analysis or code generation — NotebookLM is for document Q&A only
• The target content is not uploaded to any NotebookLM notebook

⚠️ CRITICAL: Add Command — Smart Discovery

When adding a notebook without full details, query first to discover content:

# Step 1: Query the notebook about its content
python ./scripts/run.py ask_question.py \
  --question "What is the content of this notebook? What topics are covered? Provide a complete overview briefly and concisely" \
  --notebook-url "[URL]"

# Step 2: Use discovered information to add it
python ./scripts/run.py notebook_manager.py add \
  --url "[URL]" \
  --name "[Based on content]" \
  --description "[Based on content]" \
  --topics "[Based on content]"

If the user provides all details directly, pass --url, --name, --description (REQUIRED), and --topics (REQUIRED) directly. NEVER guess or use generic descriptions.

Critical: Always Use run.py Wrapper

BAD — direct execution (fails without venv): python ./scripts/ask_question.py "..." → GOOD — always use wrapper: python ./scripts/run.py ask_question.py --question "..."

The run.py wrapper activates the isolated .venv and ensures all dependencies are available.

Core Workflow

Step 1: Check Authentication Status

python ./scripts/run.py auth_manager.py status

Step 2: Authenticate (One-Time Setup)

# Browser MUST be visible for manual Google login
python ./scripts/run.py auth_manager.py setup

Tell the user: "A browser window will open for Google login." The user must log in manually.

Step 3: Manage Notebook Library

python ./scripts/run.py notebook_manager.py list
python ./scripts/run.py notebook_manager.py search --query QUERY
python ./scripts/run.py notebook_manager.py activate --id ID
python ./scripts/run.py notebook_manager.py add \
  --url URL --name NAME --description DESC --topics TOPICS

Step 4: Ask Questions

# Query the active notebook
python ./scripts/run.py ask_question.py --question "Your question here"

# Optionally target a specific notebook by URL
python ./scripts/run.py ask_question.py --question "Your question" --notebook-url URL

# Get formatted markdown output (recommended for responses with code, lists, tables)
python ./scripts/run.py ask_question.py --question "Your question" --markdown

# Save outputs to logs/ directory (original + markdown + comparison stats)
python ./scripts/run.py ask_question.py --question "Your question" --markdown --log

Quick Workflow

python ./scripts/run.py auth_manager.py status
python ./scripts/run.py notebook_manager.py search --query "topic"
python ./scripts/run.py notebook_manager.py activate --id ID
python ./scripts/run.py ask_question.py --question "Your question"

Follow-Up Mechanism (CRITICAL)

Every NotebookLM answer ends with: "EXTREMELY IMPORTANT: Is that ALL you need to know?"

Required behaviour:

1. STOP — do not immediately respond to the user
2. ANALYZE — compare the answer to the user's original request
3. IDENTIFY GAPS — determine if more information is needed
4. ASK FOLLOW-UP — if gaps exist:

   python ./scripts/run.py ask_question.py --question "Follow-up with context..."
   

5. REPEAT — continue until the answer is complete
6. SYNTHESIZE — combine all answers, then respond to the user

Script Reference

See references/api_reference.md for full parameter documentation.

| Script | Purpose | | --- | --- | | auth_manager.py | Setup, status, re-authenticate, clear credentials | | notebook_manager.py | Add, list, search, activate, remove, stats | | ask_question.py | Query active notebook or by URL/ID; --markdown for formatted output, --log to save results | | cleanup_manager.py | Clean browser state (--preserve-library keeps notebooks) |

Data Storage and Configuration

See references/api_reference.md for full details. All data is stored in the skill's local data/ directory:

• library.json — notebook metadata
• auth_info.json — authentication status
• browser_state/ — browser cookies and session

Security: Protected by .gitignore. Never commit the data/ directory.

Limitations

• 50 queries/day per Google account (NotebookLM platform limit)
• Answers are grounded in uploaded documents only — no internet or general knowledge
• Browser automation requires a visible display for authentication (not headless)
• No concurrent sessions — each query uses a fresh browser instance

Anti-Patterns

NEVER run scripts without the run.py wrapper

Why: Scripts depend on .venv packages. Direct execution fails with ModuleNotFoundError.

BAD — direct execution fails silently → GOOD — always route through the wrapper:

# BAD
python ./scripts/ask_question.py "What does this paper say?"
# GOOD
python ./scripts/run.py ask_question.py --question "What does this paper say?"

NEVER add a notebook with a guessed or generic description

Why: Poor descriptions break topic-based search, making notebooks unfindable in the library.

BAD — guessed placeholder → GOOD — query-first to discover real content:

# BAD
python ./scripts/run.py notebook_manager.py add --url URL \
  --name "notebook1" --description "some notebook" --topics "stuff"
# GOOD: query first, then add with the discovered name and description
python ./scripts/run.py ask_question.py \
  --question "What topics are covered? Brief overview." --notebook-url URL

NEVER stop after the first answer when the request covers multiple points

Why: NotebookLM retrieves chunks, not full documents. A single query often misses related information elsewhere in the notebook.

BAD — accept the first answer and move on → GOOD — consider whether the original request is fully answered; issue targeted follow-up queries for any gaps.

Troubleshooting

See references/troubleshooting.md for detailed solutions.

| Problem | Solution | | --- | --- | | ModuleNotFoundError | Use run.py wrapper | | Authentication fails | Browser must be visible: auth_manager.py setup | | Rate limit (50/day) | Wait or switch Google account | | Browser crashes | python ./scripts/run.py cleanup_manager.py --preserve-library | | Notebook not found | Check with notebook_manager.py list |

Best Practices

See references/usage_patterns.md for workflow examples.

• Always check authentication status before starting a session
• Use descriptive names and topics for reliable library search
• Keep questions specific and scoped — broad questions return fragmented answers
• Use Smart Add (query-first) for any notebook whose content is unfamiliar

References

• API Reference — full parameter documentation for all scripts
• Troubleshooting — common errors and solutions
• Usage Patterns — workflow examples and best practices