Notebooklm Studio

NotebookLM Studio

Overview

Use this skill when the user wants Codex to turn files, URLs, YouTube links, notes, transcripts, Drive-style sources, or research material into NotebookLM notebooks and NotebookLM Studio outputs. The goal is to make Codex act as the operator: prepare sources, create or reuse notebooks, generate official artifacts where available, download outputs, and explain limitations clearly.

This skill relies on the notebooklm CLI from notebooklm-py when available. That CLI is unofficial and uses undocumented NotebookLM endpoints, so every workflow must probe the environment before assuming it can run.

Operating Rules

1. Only ingest sources the user owns, provided, or is allowed to use. Do not bypass paywalls, DRM, private access controls, or copyright restrictions.
2. Prefer official NotebookLM outputs over local reconstructions. If an output is not exposed by the CLI or account plan, create a source-grounded fallback and label it as derived.
3. Keep artifacts traceable: return notebook title/id when known, source count, generated artifact names, local download paths, and any failed/skipped outputs.
4. For current feature availability, limits, and export formats, load references/notebooklm-capabilities.md.
5. For exact CLI commands and supported automation flags, load references/notebooklm-py-cli.md.
6. For mixed source handling and "any source" requests, load references/source-ingestion.md.
7. For output mapping and fallback decisions, load references/artifact-strategy.md.

First-Time Initialization

When a new user asks to initialize, set up, or start using this skill, make the setup as automatic as possible. The user should not be asked to manually configure package paths, folders, profiles, or commands unless an error requires it.

1. Run the bootstrap check:

python scripts/bootstrap_notebooklm.py --json --print-guide

2. If notebooklm is missing, ask for permission to install dependencies, then run:

python scripts/bootstrap_notebooklm.py --install --print-guide

3. If authentication is missing or expired, ask the user to complete the Google login step, then run:

python scripts/bootstrap_notebooklm.py --login --auth-test --print-guide

The login browser is the manual boundary: Codex can start it, but the user must choose the Google account, pass any MFA, and grant access. After login returns, Codex should run the environment check again and tell the user whether NotebookLM is ready.

4. Immediately after initialization, respond with a concise but complete guide. Prefer the generated guide from bootstrap_notebooklm.py --print-guide; load references/quickstart-guide.md if a fuller tutorial is useful.

5. If the user wants the local control room UI, start the dashboard from the current project/workspace:

python scripts/dashboard_server.py --host 127.0.0.1 --port 8765 --profile default --out-dir ./notebooklm_outputs/dashboard

Then give the user the local URL. If port 8765 is occupied, choose the next available local port and report the actual URL.

Standard Workflow

1. Clarify the deliverable only if needed: audience, language, artifact types, output folder, and whether to create a new notebook or reuse an existing one.
2. Run an environment check:

python scripts/validate_environment.py --json

If notebooklm is missing, explain that NotebookLM automation needs notebooklm-py plus Google authentication. Ask before installing packages or opening browser login.

3. Build a source manifest before uploading:

python scripts/source_manifest.py --output /tmp/notebooklm-sources.json SOURCE...

Review warnings for unsupported extensions, over-large files, too many sources, missing files, paywalled URLs, or YouTube videos without usable captions.

4. Create a dry-run plan with the optimized artifact pipeline:

python scripts/artifact_pipeline.py plan --notebook-title "Project Title" --source SOURCE --artifact audio --artifact video --artifact mind-map --artifact slide-deck --download --out-dir ./outputs

5. Run only after the plan looks correct and authentication is ready:

python scripts/artifact_pipeline.py run --notebook-title "Project Title" --source SOURCE --artifact audio --artifact video --artifact mind-map --artifact slide-deck --download --out-dir ./outputs --status-file ./outputs/notebooklm-jobs.json

6. Verify outputs: list artifacts, check local files exist, inspect file sizes, and summarize what was generated.

Optimized Artifact Pipeline

For multi-artifact requests, prefer scripts/artifact_pipeline.py over the older linear orchestrator. It avoids the "audio must finish before video starts" problem by submitting requested artifacts with --no-wait where NotebookLM supports it, then polling a shared job table. Completed artifacts are downloaded immediately; failed artifacts do not block unrelated outputs.

Use this pattern for requests like "make audio, video, mind map, and slides":

python scripts/artifact_pipeline.py run \
  --notebook-title "Research Pack" \
  --source ./report.pdf \
  --source https://example.com/article \
  --artifact audio \
  --artifact video \
  --artifact mind-map \
  --artifact slide-deck \
  --download \
  --download-slide-format both \
  --convert-mind-map-html \
  --status-file ./notebooklm_outputs/research-pack-jobs.json \
  --out-dir ./notebooklm_outputs

The pipeline should report each artifact as planned, submitted, in_progress, completed, downloaded, failed, submit_failed, download_failed, or timeout. Do not let a failed video prevent audio, mind-map, or slide-deck delivery.