notebooklm-research

name	notebooklm-research
description	Research a topic with NotebookLM — create notebook, run deep research, auto-import sources, and summarize. Checks notebooklm installation, login, and skill installation first.

NotebookLM Research Workflow

Research a topic using NotebookLM: create a notebook, run deep web research with auto-import sources, and summarize.

Core Principle: Non-Blocking

Never use notebooklm research wait, notebooklm source wait, or notebooklm artifact wait in main conversation. These commands block for minutes. Instead:

1. Launch the operation (e.g., --no-wait flag)
2. Immediately return to the user — tell them what's running and how long it'll take
3. The user will check back later and re-invoke if needed

Workflow Steps

When this skill is invoked, follow these steps in order:

Prerequisites: Check Environment

Before starting, verify prerequisites:

1. Check NotebookLM Installation

   notebooklm --version 2>/dev/null || pip3 show notebooklm-py 2>/dev/null
   

   If not installed:

   pip install notebooklm-py
   notebooklm skill install
   

2. Check Login Status

   notebooklm status 2>&1
   

   If authentication fails or session expired:

   notebooklm auth check 2>&1
   

   If not logged in, tell the user to run notebooklm login first and stop.

3. Check NotebookLM Skill Installation

   ls ~/.claude/skills/notebooklm/SKILL.md 2>/dev/null && echo "Skill installed"
   

   If not installed:

   notebooklm skill install
   

Step 1: Create Notebook & Start Research

1. If no topic is provided, ask the user: "What research topic would you like to explore?"

2. Create a new notebook with the research topic as title:

   notebooklm create "<research-title>" --json
   

   Parse the notebook ID from the JSON output.

3. Set the notebook as current context:

   notebooklm use <notebook-id>
   

4. Start deep web research (non-blocking):

   notebooklm source add-research "<research-query>" --mode deep --no-wait
   

5. Immediately return to the user with:

   • Notebook title and ID
   • Research has started in background (deep mode takes ~3-5 minutes)
   • Tell the user to wait and then come back to check progress and import sources
   • The user can later run notebooklm research status -n <notebook-id> to check

Step 2: Check Research Status (User Returns)

When the user comes back to check:

1. Check research status:

   notebooklm research status -n <notebook-id>
   

2. If research is still in progress → tell the user it's still running, wait a bit more.

3. If research is complete → show the found sources and ask the user whether to proceed with importing.

Step 3: Import Sources (User Confirms)

1. Do NOT run research wait --import-all in main conversation. Instead:

   • Tell the user this step takes a few minutes
   • Suggest running: notebooklm research wait --import-all -n <notebook-id> and waiting for it to complete
   • The user can come back after running it

2. Alternatively, check if sources are already imported:

   notebooklm source list -n <notebook-id>
   

   If sources are already there and ready → proceed to Step 5.

Step 4: Wait for Source Processing (User-Initiated)

• Tell the user that after importing, sources need ~1-2 minutes to process
• User can check with: notebooklm source list -n <notebook-id> (look for "ready" status)
• Come back when all key sources show "ready"

Step 5: Show General Description

Ask NotebookLM to generate a general description/summary of the notebook content:

notebooklm ask "请用中文概述这个 Notebook 的核心内容和主要发现" -n <notebook-id>

Present the generated summary to the user.

Example

User: /notebooklm-research Russian space program 2026

Agent:

1. ✅ Checks notebooklm installed (v0.4.1)
2. ✅ Verifies login status (authenticated)
3. ✅ Confirms skill installed
4. ✅ Creates notebook "Russian space program 2026", starts deep research
5. ✅ Returns immediately — tells user "Research started (~5 min), come back later to check"
6. 🔄 User returns → agent checks status → shows found sources → asks user to import
7. 🔄 User imports → comes back → agent verifies and generates summary