An automated browser-driven skill that bridges Claude Code and Google NotebookLM to query your documents and retrieve citation-backed Gemini answers.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install phy-notebooklm
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install phy-notebooklm using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is NotebookLM Research Assistant?
The NotebookLM Research Assistant Skill bridges the gap between your local AI development agent and Google's NotebookLM, allowing you to run document-grounded, zero-hallucination queries directly from Claude Code. By integrating browser automation into your development cycles, this tool automates authentication, manages document libraries, and uses Gemini's context windows to answer queries solely from your uploaded sources. Integrating Openclaw Skills like this helps developers query technical APIs, textbook folders, or research files on the fly.
Built to operate seamlessly behind the scenes, the skill features a managed execution environment that eliminates virtual environment mismatches. Whether querying system architecture specifications or indexing active course syllabus repositories, this assistant streamlines complex technical document ingestion.
NotebookLM Research Assistant Use Cases
- Activated when you explicitly request Claude to "check my notebooks", "query my docs", or ask a question directed at a NotebookLM notebook.
- Triggered immediately when a Google NotebookLM link (
https://notebooklm.google.com/notebook/...) is shared in your agent's command interface. - Used to execute batch file uploads (PDFs, DOCX, TXT, MD) into active Google notebooks for course training or library compilation.
- Highly valuable for technical teams needing strictly grounded codebase audits without passing intellectual property to ungrounded LLMs.
How NotebookLM Research Assistant Works
- Prompt Parsing: The agent detects a NotebookLM-relevant command or URL and initializes the
run.pywrapper script. - Authentication Audit: The skill checks status via
auth_manager.pyto determine if a valid Chromium session state is available locally. - Environment Sync: If necessary, it handles environment isolation, updates dependencies, and provisions the correct browser binary.
- Browser Automation Interface: The browser starts (headless by default, visible during setup), navigates to the targeted notebook URL, types the query, and waits for output generation.
- Recursive Follow-Up Loop: The engine detects finalization prompts, evaluates if supplementary data is needed, automatically executes secondary queries, and synthesizes the full answer before returning it.
NotebookLM Research Assistant Setup
To setup the NotebookLM integration via Openclaw Skills, run the configuration and login sequences:
1. Initial One-Time Authentication
Run the setup command. Note that a visible browser will launch automatically for manual Google login:
python scripts/run.py auth_manager.py setup
Verify authentication is successful:
python scripts/run.py auth_manager.py status
2. Registering and Organizing Notebooks
To add a notebook to your index, query it automatically to extract metadata (Smart Add), or add it manually:
# Smart Add (Recommended auto-discovery)
python scripts/run.py ask_question.py --question "What topics are covered in this notebook? Provide a brief overview." --notebook-url "[YOUR_NOTEBOOK_URL]"
# Manual Add (Requires metadata fields)
python scripts/run.py notebook_manager.py add \
--url "https://notebooklm.google.com/notebook/your-notebook-uuid" \
--name "Microservices Architecture Docs" \
--description "Database schemas and system design guidelines for scaling microservices" \
--topics "architecture,databases,devops"
3. Querying the Assistant
Ask questions using either notebook identifiers or direct URLs:
# Query by active or targeted notebook ID
python scripts/run.py ask_question.py --question "What is the backoff policy for database connections?" --notebook-id notebook-uuid
NotebookLM Research Assistant Data Schema & Taxonomy
All session assets and configuration parameters are localized within ~/.claude/skills/notebooklm/data/:
| Directory/File | Data Scope | Security & Git Status |
|---|---|---|
library.json |
Local catalog listing notebook configurations, tags, descriptions, and IDs. | Git-ignored, confidential |
auth_info.json |
Authentication checks and expiration details for persistent browser states. | Git-ignored, confidential |
browser_state/ |
Chromium cookies, cached profiles, and storage footprints. | Git-ignored, confidential |
Configuration Options (.env)
Adjust optional runtime behaviors inside the skill root folder:
HEADLESS=false # Toggle browser execution mode (true/false)
STEALTH_ENABLED=true # Emulates natural human interactions to skip detection
TYPING_WPM_MIN=160 # Lower boundary for text generation speed
TYPING_WPM_MAX=240 # Upper boundary for text generation speed
DEFAULT_NOTEBOOK_ID= # Default target if no ID/URL is parsed
NotebookLM Research Assistant Advanced Features
- Contextual Follow-Up Loops: Automatically identifies if the received answer satisfies the query context; if not, triggers recursive prompts behind the scenes.
- Metadata Auto-Discovery: Executes Smart Add queries to extract descriptive topics from unregistered notebooks prior to cataloging.
- Stealth Automation Controls: Implements randomized typing speeds (WPM offsets) and human-like cursor movements to guarantee flawless navigation.
- Automated Directory Uploader (
upload_files.py): Batches PDF, DOCX, TXT, and Markdown files to specific notebooks complete with dry-run evaluation frameworks. - Environment Safety Nets: Relies on a unified
run.pyscript wrapper standard across Openclaw Skills to handle.venvscaffolding automatically.
SKILL.md
Loading
Related Openclaw Skills
Accessibility (A11y) Reviewer
An AI-powered web accessibility reviewer that automatically audits web pages and code structures against WCAG 2.2 AA compliance standards.
UI Reviewer
An AI-powered frontend code reviewer that automates design system compliance, structural audits, responsiveness checks, and visual consistency for React and Vue applications.
Project Truth Audit
A comprehensive whole-repository audit skill for AI agents to contrast stated system claims with the underlying reality of files, tests, and runtime evidence.
Alibaba Cloud Lindorm Vector Database Migration Assistant
An automated AI agent skill for seamless vector database migration from Milvus, Elasticsearch, Qdrant, or CSV to Alibaba Cloud Lindorm with built-in checkpoint resume.
Deep Research
A structured, parallel-first research skill that enables AI agents to decompose complex queries, execute concurrent searches, and synthesize high-confidence answers.
Frontend Design
An expert frontend design skill that guides AI agents to construct high-quality, production-grade web interfaces and components that avoid generic AI aesthetics.
Featured*
