NotebookLM

Use this skill to let Codex query Google NotebookLM and return answers grounded in the user's uploaded documents.

This skill is for local Codex environments that can:

• run shell commands
• install Python dependencies
• access the network
• open a visible browser window for Google login

Use bundled scripts as black boxes

Always use the wrapper:

python scripts/run.py <script> [args...]

Do not call scripts in scripts/ directly unless you are debugging the skill itself. The wrapper creates .venv, installs dependencies, and runs the script with the correct interpreter.

If the first setup is interrupted, rerun the same wrapper command. The wrapper now repairs incomplete .venv directories automatically.

When to use

Trigger this skill when the user:

• mentions NotebookLM
• shares a NotebookLM URL
• asks to query their docs through NotebookLM
• wants to save or manage NotebookLM notebooks locally
• asks for answers grounded only in uploaded notebook sources

Core workflow

1. Check authentication first

python scripts/run.py auth_manager.py status

If authentication is missing or stale, run setup.

2. Authenticate with Google

python scripts/run.py auth_manager.py setup

Important:

• the browser must be visible for login
• tell the user a browser window will open
• wait for them to complete Google login manually

3. Resolve the notebook

List saved notebooks:

python scripts/run.py notebook_manager.py list

If the user provided a notebook URL that is not in the library yet, prefer Smart Add:

python scripts/run.py ask_question.py \
  --question "What is the content of this notebook? What topics are covered? Provide a brief overview." \
  --notebook-url "[URL]"

python scripts/run.py notebook_manager.py add \
  --url "[URL]" \
  --name "[Specific notebook name]" \
  --description "[Specific description based on notebook content]" \
  --topics "[topic1,topic2,topic3]"

Do not invent generic metadata if the notebook content is still unknown.

4. Ask the notebook

Use one of:

python scripts/run.py ask_question.py --question "Your question here"
python scripts/run.py ask_question.py --question "..." --notebook-id notebook-id
python scripts/run.py ask_question.py --question "..." --notebook-url "https://notebooklm.google.com/notebook/..."

Add --show-browser only when debugging UI failures.

Follow-up rule

ask_question.py appends a follow-up reminder to every answer. Treat that as an internal prompt to keep researching before answering the user.

Before you respond:

1. Compare the NotebookLM answer to the user's actual request.
2. Identify missing details, edge cases, or implementation gaps.
3. Ask one or more follow-up questions if the first answer is incomplete.
4. Synthesize the final response only after the information is sufficient.

Each NotebookLM query is stateless, so follow-up questions must include the needed context.

Command reference

Read commands.md when you need the exact CLI for:

• authentication
• library management
• notebook queries
• cleanup

Troubleshooting

Read troubleshooting-codex.md when:

• login fails
• the browser does not launch
• selectors break after a NotebookLM UI change
• Patchright or Chrome installation fails

Operational notes

• Skill data is stored under the skill directory in data/.
• First use creates .venv in the skill directory.
• Environment setup installs Python dependencies and Patchright's Chrome runtime.
• Dependency installation streams pip output so DNS or mirror failures are visible instead of appearing stuck.
• NOTEBOOKLM_PIP_INDEX_URL can be used to point pip at a reachable package mirror.
• Authentication uses a persistent browser profile plus saved storage state.
• Free Google accounts may hit NotebookLM rate limits.

Limits

• Each query opens a fresh browser session.
• The user must already have a NotebookLM notebook with uploaded sources.
• Browser automation may break if NotebookLM changes its UI.