NotebookLM Studio Skill

中文 | English

NotebookLM Studio Skill

AI agent skill for Google NotebookLM — import sources (URLs, YouTube, files, text) and generate podcasts, videos, reports, quizzes, flashcards, mind maps, slides, infographics, and data tables.

Works with Claude Code, OpenClaw, Codex, and any agent that supports the Agent Skills specification.

Features

9 artifact types: audio, video, report, quiz, flashcards, mind-map, slide-deck, infographic, data-table
All source types: URLs, YouTube, text notes, PDF, Word, audio, images, Google Drive
Multi-language output: generates content in 30+ languages (default: Traditional Chinese)
Cross-platform: works with any AI agent that reads SKILL.md
Telegram delivery: two-round delivery with real-time status tracking (via OpenClaw)
Audio compression: ffmpeg post-processing for Telegram's 50MB file size limit
CLI-driven: uses notebooklm CLI directly — no custom Python wrappers

Usage

Just tell your AI agent what you want. No CLI knowledge needed.

Generate a study package from a URL:

Generate a report, quiz, and flashcards from this article https://example.com/deep-learning-intro

Turn a YouTube video into a podcast and slides:

Make a podcast and slide deck from this video, use debate format https://youtube.com/watch?v=dQw4w9WgXcQ

Create an infographic from a PDF:

Create a sketch-note style infographic from this paper (attach file.pdf)

Full package via Telegram (OpenClaw):

幫我用這篇文章做一份完整的學習包，包含 report、quiz、flashcards、podcast 和簡報 https://example.com/article

The agent handles everything: creating the notebook, adding sources, generating artifacts, and delivering results back to you.

How It Works

You ──────── "Generate a report and slides from this URL"
                │
                ▼
        AI Agent (Claude Code / OpenClaw / Codex)
                │  reads SKILL.md → 9-step workflow
                ▼
        notebooklm CLI
                │  create notebook → add sources → generate → download
                ▼
        ./output/<topic>/
            ├── report.md
            ├── slides.pdf
            ├── podcast.mp3
            └── ...
                │
                ▼  (optional, OpenClaw)
        Telegram ── delivers each artifact as it completes

Artifact Types

Type	CLI Command	Est. Time	Output
Audio (Podcast)	`generate audio`	5-30 min	MP3
Video	`generate video`	5-30 min	MP4
Report	`generate report`	1-2 min	Markdown
Quiz	`generate quiz`	1-2 min	JSON/MD/HTML
Flashcards	`generate flashcards`	1-2 min	JSON/MD/HTML
Mind Map	`generate mind-map`	Instant	JSON
Slide Deck	`generate slide-deck`	2-10 min	PDF/PPTX
Infographic	`generate infographic`	2-5 min	PNG
Data Table	`generate data-table`	1-2 min	CSV

See references/artifacts.md for full CLI options per artifact type.

Setup

Prerequisites

Python 3.10+
ffmpeg (for audio compression)
OS: macOS, Linux (Ubuntu 20.04+), or Windows

1. Install the skill

Option A — ClawHub (recommended):

npm i -g clawhub        # install ClawHub CLI (one-time)
clawhub install notebooklm-studio

Option B — Git clone:

git clone --recurse-submodules https://github.com/jasontsaicc/notebooklm-studio-skill.git
cd notebooklm-studio-skill

2. Install notebooklm CLI

Both Option A and B require steps 2–5 below.

The skill requires the notebooklm CLI tool. This is a separate dependency regardless of how you installed the skill.

pip install "notebooklm-py[browser]"
playwright install chromium

Ubuntu/Debian — also install system dependencies for Chromium:

playwright install-deps chromium

3. Install ffmpeg

macOS:

brew install ffmpeg

Ubuntu/Debian:

sudo apt update && sudo apt install -y ffmpeg

4. Authenticate

On a machine with a browser (Mac/Windows/Linux desktop):

notebooklm login

On a headless server (e.g. Ubuntu VPS):

Run login on your local machine first, then transfer the credential:

# Local machine — login and verify
notebooklm login
notebooklm auth check

# Transfer to server
ssh user@server "mkdir -p ~/.notebooklm"
scp ~/.notebooklm/storage_state.json user@server:~/.notebooklm/storage_state.json
ssh user@server "chmod 600 ~/.notebooklm/storage_state.json"

5. Verify

notebooklm auth check --test

Expected: all checks pass, token fetch succeeds.

6. Install as agent skill (Option B only)

Skip this step if you used ClawHub. For Git clone:

Claude Code:

ln -s "$(pwd)" ~/.claude/skills/notebooklm-studio

OpenClaw:

ln -s "$(pwd)" /path/to/openclaw/skills/notebooklm-studio

Other agents: Place or symlink this directory where your agent discovers skills.

Quick Demo (CLI)

After setup, try a quick end-to-end test without an AI agent:

# Create a notebook
notebooklm create "Test Notebook $(date +%Y%m%d)"
notebooklm use <notebook_id>    # use the ID from output above

# Add a source
notebooklm source add "https://en.wikipedia.org/wiki/Feynman_technique"

# Generate a report (fastest artifact, ~1 min)
notebooklm generate report --format study-guide --wait

# Download
mkdir -p output/feynman-technique
notebooklm download report ./output/feynman-technique/report.md

# Check result
cat ./output/feynman-technique/report.md

If this works, your setup is complete. The AI agent will follow the same workflow automatically via SKILL.md.

Repository Structure

notebooklm-studio-skill/
├── SKILL.md                         # Agent skill definition (9-step workflow)
├── README.md
├── LICENSE
├── notebooklm-py/                   # git submodule (notebooklm CLI)
├── references/
│   ├── artifacts.md                 # 9 artifact types + CLI options
│   ├── artifact-options.md          # ASK/OFFER/SILENT option priorities
│   ├── source-types.md              # Source types & detection rules
│   ├── output-contracts.md          # Output format specifications
│   └── telegram-delivery.md         # Telegram delivery contract
├── scripts/
│   ├── compress_audio.sh            # ffmpeg audio compression
│   └── recover_tier2_delivery.sh    # Cron recovery for Tier 2 artifacts
└── assets/                          # Screenshots and demo media

OpenClaw Timeout & Recovery Setup

Tier 2 artifacts (podcast, video, slides) take 5–40 minutes to generate. If the agent times out before they finish, they'll be generated on NotebookLM's server but never downloaded or delivered. This section sets up automatic recovery.

1. Agent Timeout

Set timeoutSeconds: 1800 (30 min) for the notebooklm agent in your OpenClaw config:

{
  "agents": {
    "list": [
      {
        "id": "notebooklm",
        "timeoutSeconds": 1800
      }
    ]
  }
}

This covers most Tier 2 artifacts. For cinematic video (Veo 3, ~40 min), consider 2400.

2. Tier 2 Recovery Cron (every 5 min)

If the agent times out mid-delivery, delivery-status.json tracks what's still pending. The recovery script polls, downloads, and updates status automatically.

# crontab -e
*/5 * * * * cd /path/to/notebooklm-studio-skill && bash scripts/recover_tier2_delivery.sh ./output >> /var/log/notebooklm-recovery.log 2>&1

What it does:

Scans output/*/delivery-status.json for "status": "pending" artifacts
Polls each via notebooklm artifact poll <task_id>
If completed → downloads, compresses audio, updates status
If failed → marks as failed
If processing → skips (next run will retry)

3. Health Check Cron (every 30 min)

Catch expired sessions before they block recovery:

# crontab -e
*/30 * * * * notebooklm auth check --test --json | python3 -c "import json,sys; d=json.load(sys.stdin); exit(0 if d['status']=='ok' else 1)" || echo "$(date): AUTH EXPIRED — run notebooklm login" >> /var/log/notebooklm-health.log

Troubleshooting

"Tier 2 artifacts generated but never delivered"

This happens when the agent times out during artifact wait. Check and recover:

# See what's pending
python3 -c "
import json, glob
for f in glob.glob('output/*/delivery-status.json'):
    data = json.load(open(f))
    pending = [a for a in data['artifacts'] if a['status'] == 'pending']
    if pending:
        print(f'{f}: {len(pending)} pending')
        for a in pending:
            print(f'  {a[\"type\"]} — task_id: {a[\"task_id\"]}')
"

# Manual recovery
bash scripts/recover_tier2_delivery.sh ./output

"Auth check failed" in recovery log

Session expired. Re-login and recovery will resume on the next cron cycle:

notebooklm login
notebooklm auth check --test   # verify

"Recovery runs but artifacts stay pending"

The artifact may still be generating. Check manually:

notebooklm artifact poll <task_id> --json

If processing → wait. If stuck for 60+ minutes → check the NotebookLM web UI directly.

Updating notebooklm-py

cd notebooklm-py && git pull origin main && cd ..
pip install -e "notebooklm-py[browser]"

Powered By

notebooklm-py — Unofficial Python API & CLI for Google NotebookLM

License

MIT

Name		Name	Last commit message	Last commit date
Latest commit History 35 Commits
.claude		.claude
assets		assets
notebooklm-py @ 8f9dd3a		notebooklm-py @ 8f9dd3a
references		references
scripts		scripts
.clawignore		.clawignore
.gitignore		.gitignore
.gitmodules		.gitmodules
CLAUDE.md		CLAUDE.md
LICENSE		LICENSE
README.md		README.md
README.zh-TW.md		README.zh-TW.md
SKILL.md		SKILL.md

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

NotebookLM Studio Skill

Features

Usage

How It Works

Artifact Types

Setup

Prerequisites

1. Install the skill

2. Install notebooklm CLI

3. Install ffmpeg

4. Authenticate

5. Verify

6. Install as agent skill (Option B only)

Quick Demo (CLI)

Repository Structure

OpenClaw Timeout & Recovery Setup

1. Agent Timeout

2. Tier 2 Recovery Cron (every 5 min)

3. Health Check Cron (every 30 min)

Troubleshooting

Updating notebooklm-py

Powered By

License

About

Uh oh!

Releases

Packages

Uh oh!

Contributors

Uh oh!

Languages

Folders and files

Latest commit

History

Repository files navigation

NotebookLM Studio Skill

Features

Usage

How It Works

Artifact Types

Setup

Prerequisites

1. Install the skill

2. Install notebooklm CLI

3. Install ffmpeg

4. Authenticate

5. Verify

6. Install as agent skill (Option B only)

Quick Demo (CLI)

Repository Structure

OpenClaw Timeout & Recovery Setup

1. Agent Timeout

2. Tier 2 Recovery Cron (every 5 min)

3. Health Check Cron (every 30 min)

Troubleshooting

Updating notebooklm-py

Powered By

License

About

Topics

Resources

License

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages