notebooklm-by-jack-vidal

name

notebooklm-by-jack-vidal

description

Easy NotebookLM access for Jack Vidal's students. One-skill setup that handles install, fast authentication (instant-ENTER trick that avoids the usual blocking prompt), and common workflows for chatting with sources and generating podcasts, videos, quizzes, mind maps, and more. Activates on /notebooklm, "use notebooklm", or intent like "create a podcast about X", "summarize these sources", "make a quiz from notebook Y", "add this PDF to NotebookLM".

NotebookLM by Jack Vidal

A friendly entry point to Google NotebookLM for Jack Vidal's students. Wraps the open-source notebooklm-py CLI (https://github.com/teng-lin/notebooklm-py) with a streamlined setup, the fastest known login flow, and copy-paste workflows for the things students actually want to do.

What you get: every NotebookLM feature (sources, chat, podcasts, videos, quizzes, flashcards, mind maps, slide decks, infographics, reports, web research) plus a few things the web UI doesn't expose (batch downloads, JSON/Markdown quiz export, mind map JSON, PPTX slides).

1. First-time bootstrap (the agent runs this — students don't type anything)

On first activation, before doing anything else, the agent must verify the runtime is installed and install it if missing. Students should never be told to run install commands manually — the agent does it for them.

Step 1.1 — Check whether the CLI is already installed

Run this silently (don't ask the student first — it's read-only):

notebooklm --version

• Exit code 0 with a version string → Skip to section 2 (Authentication). Do not reinstall.
• "command not found" / non-zero exit → Continue with Step 1.2.

Step 1.2 — Install the CLI

Tell the student: "First time using this — installing the NotebookLM runtime (~1 minute)." Then run (this is the one place to ask permission, since it modifies the Python environment):

python -m pip install "notebooklm-py[browser]"

Notes for the agent:

• Use python -m pip not bare pip — survives PATH issues across Windows/macOS/Linux.
• If python isn't found, try python3 -m pip. If neither works, stop and tell the student: "Python 3.10+ is required. Install from https://python.org/downloads then re-run."
• If pip warns about externally-managed environment (Debian/Ubuntu/macOS Homebrew Python), retry with python -m pip install --user "notebooklm-py[browser]".

Step 1.3 — Install the Chromium browser used for login

Run (no student input needed — this is the runtime the auth flow will use):

playwright install chromium

Tell the student: "Downloading Chromium (~330 MB, takes 1–2 minutes on a normal connection)." This step is required even if they have Chrome installed — Playwright uses its own bundled build.

Step 1.4 — Verify install succeeded

notebooklm --version

Should now print a version. If it doesn't, the entry point script isn't on PATH — fall back to invoking the module directly: python -m notebooklm.cli (substitute that for notebooklm in everything that follows). Tell the student to restart their terminal once after install so the PATH update takes effect for future sessions.

Then continue to section 2.

2. Fast authentication

The standard notebooklm login opens a browser, waits for you to sign in, then blocks forever waiting for you to press ENTER in the terminal. That works fine when you're typing manually, but it breaks every time an AI agent tries to drive it. Use this pattern instead — it pre-queues the ENTER so authentication finishes the moment your Google session is in the browser profile.

The decision tree

Always start with this:

notebooklm auth check --test

Then branch on the result:

Result	Run this
Authentication is valid	Done — skip to step 3.
Storage file not found but you've signed in before	Path A below (instant, ~2 seconds)
First time ever, or Token fetch ✗ fail (cookies expired)	Path B below (one-time interactive, ~30–90s)

Path A — Instant re-extract

The Chromium profile on disk still has your Google session. Just dump it to storage_state.json:

echo "" | notebooklm login

That's it. The pipe sends ENTER immediately, the script writes the cookie file, done.

Path B — Interactive sign-in (first time)

You'll sign into Google in a Chromium window. The trick is to launch it with a stdin pipe that auto-presses ENTER on a timer, so the script can finish even if no human is at the keyboard.

If a human is at the keyboard (the simple case):

notebooklm login --fresh
# 1. Sign in to Google in the Chromium window
# 2. Wait until you see your NotebookLM home page
# 3. Switch back to the terminal and press ENTER

If an AI agent is driving the install (the magic case):

Run this in the agent's background-task mode so the agent doesn't block:

( sleep 240; echo "" ) | notebooklm login --fresh

The agent tells the student "sign in within 4 minutes." When the timer fires, ENTER is sent automatically and the cookie file is written. Even faster: as soon as the student says "I'm signed in," the agent kills the background task and runs Path A — that captures the live session in ~2 seconds instead of waiting out the timer.

Verify

notebooklm auth check --test    # all five rows should be ✓ pass
notebooklm list                 # should show your notebooks

3. Everyday commands

Want to...	Command
See your notebooks	notebooklm list
Pick a notebook to work in	notebooklm use <notebook_id>
Create a new notebook	notebooklm create "Title"
Add a web page	notebooklm source add "https://..."
Add a PDF / Word / Markdown / EPUB / audio / video / image	notebooklm source add ./file.pdf
Add a YouTube video	notebooklm source add "https://youtube.com/watch?v=..."
Have AI do web research and import sources	notebooklm source add-research "your topic"
List sources in current notebook	notebooklm source list
Ask a question about your sources	notebooklm ask "your question"
Get an answer with citations	notebooklm ask "your question" --json
Save the answer as a notebook note	notebooklm ask "..." --save-as-note
Show conversation history	notebooklm history

4. Generating things

All of these run for several minutes. Start them, capture the artifact ID, and check back later — don't sit and wait.

Want to make a...	Command	Then download with
Podcast (audio overview)	notebooklm generate audio "make it engaging"	notebooklm download audio ./podcast.mp3
Video overview	notebooklm generate video --style whiteboard	notebooklm download video ./video.mp4
Slide deck	notebooklm generate slide-deck	notebooklm download slide-deck ./slides.pdf (or .pptx --format pptx)
Quiz	notebooklm generate quiz --difficulty medium	notebooklm download quiz ./quiz.json (or .md --format markdown)
Flashcards	notebooklm generate flashcards	notebooklm download flashcards ./cards.json
Mind map	notebooklm generate mind-map	notebooklm download mind-map ./map.json
Briefing doc / study guide / blog post	notebooklm generate report --format study-guide	notebooklm download report ./report.md
Data table from sources	notebooklm generate data-table "compare key concepts"	notebooklm download data-table ./data.csv
Infographic	notebooklm generate infographic --orientation portrait	notebooklm download infographic ./image.png

To check progress: notebooklm artifact list. To wait for one in a script: notebooklm artifact wait <artifact_id>.

5. Activation triggers

Invoke this skill when the student says any of:

• /notebooklm, "use notebooklm", any direct mention of NotebookLM
• "Create a podcast / audio overview about [topic]"
• "Summarize these URLs / PDFs / sources"
• "Make a quiz / flashcards / mind map / video / slide deck / infographic from notebook X"
• "Add this PDF / link / YouTube video to NotebookLM"
• "Run web research on [topic] and import the sources"
• "Download the podcast / quiz / report from notebook X"
• "What does my notebook say about [topic]?"

6. Autonomy rules (for the AI agent using this skill)

Run automatically — no need to ask:

• notebooklm --version (silent install check)
• playwright install chromium (after pip install — paired bootstrap step)
• auth check, auth check --test, status, list, source list, artifact list, language list, language get, history
• notebooklm use <id>, create, source add, source add-research (fast mode)
• ask "..." without --save-as-note
• The Path A/B login flow above when auth check --test fails

Tell the student before running, but don't wait for confirmation:

• python -m pip install "notebooklm-py[browser]" (modifies Python env, but is the explicit purpose of installing this skill — students who installed the skill expect this)

Confirm with the student first:

• notebooklm delete / notebook delete — destructive
• notebooklm generate * — runs for many minutes, can fail with rate limits
• notebooklm download * — writes to filesystem
• artifact wait / source wait / research wait in the main conversation — blocks for minutes
• ask "..." --save-as-note, history --save — writes notes into the notebook
• language set — changes a global account-wide setting

7. Workflows worth memorizing

"Turn this article into a podcast" (5–10 min)

1. notebooklm create "Podcast: <topic>" (or use an existing notebook)
2. notebooklm source add "<url-or-path>"
3. Wait ~30–60s for processing: notebooklm source list until status is ready
4. notebooklm generate audio "Focus on the most surprising insight" --json → note the task_id
5. Come back in 10–20 min, notebooklm artifact list
6. notebooklm download audio ./podcast.mp3

"Study these PDFs" (1–2 min)

1. notebooklm create "Studying: <subject>"
2. notebooklm source add ./chapter1.pdf (repeat for each file)
3. notebooklm generate flashcards --difficulty medium
4. notebooklm generate quiz
5. notebooklm ask "Quiz me on the hardest concepts"

"Research a topic from scratch" (2–5 min for fast, 15–30 min for deep)

1. notebooklm create "Research: <topic>"
2. notebooklm source add-research "<topic query>" --mode fast (or --mode deep --no-wait)
3. Once sources are imported, notebooklm generate report --format briefing-doc
4. notebooklm download report ./briefing.md

8. Useful flags

• --json — machine-readable output for any command. Pipe to jq.
• -n <notebook_id> / --notebook <notebook_id> — target a specific notebook without use. Required when running multiple agents in parallel.
• -s <source_id> — limit a question or generation to specific sources.
• --language <code> — output language for generated content (en, es, fr, ja, zh_Hans, he, etc. — notebooklm language list for all).
• --retry N — auto-retry on rate limits with exponential backoff.

9. Troubleshooting

Symptom	Try this
Login hangs forever	You forgot the echo "" | prefix. Cancel and re-run with the pipe.
Chromium opens then immediately closes	Stale browser profile. Run notebooklm login --fresh (Path B).
auth check says cookies expired	Try Path A first. If still failing, run Path B with --fresh.
Generation fails with rate-limit error	Wait 5–10 minutes and retry, or use the NotebookLM web UI as a fallback.
"No notebook context" error	You forgot to set context. Pass -n <notebook_id> or run notebooklm use <id>.
Anything else weird	notebooklm doctor — runs full environment health check.

10. Where things live

• Auth cookies: ~/.notebooklm/profiles/default/storage_state.json
• Browser profile: ~/.notebooklm/profiles/default/browser_profile/
• Notebook context: ~/.notebooklm/profiles/default/context.json
• Multiple Google accounts: notebooklm profile create work then notebooklm -p work login

11. Credits and the deeper reference

This skill is built on notebooklm-py by Teng Lin (https://github.com/teng-lin/notebooklm-py, MIT). It is an unofficial library that uses undocumented Google APIs — APIs may change without notice, rate limits apply, and Google can throttle heavy use. Best for personal learning, research, and study workflows.

For every command, every flag, every artifact format, JSON schemas, and the full troubleshooting table, see the upstream skill at ~/.claude/skills/notebooklm/SKILL.md (installed by notebooklm skill install). This skill is the friendly front door; that one is the encyclopedia.