NotebookLM Auth Bypass

Programmatic NotebookLM control with auto-recovery for authentication errors.

Security Scan

The skill's behavior (automatically extracting browser cookies and writing them to disk and to a persistent Windows environment variable) matches its stated purpose, but it asks for and performs sensitive actions (stealing session cookies and modifying user environment) that are high-risk and deserve extra scrutiny before consenting.

Purpose & Capability

The name/description (recover NotebookLM authentication) aligns with the included script: it automates a persistent Playwright browser profile, extracts storage_state cookies, and applies them to NOTEBOOKLM_AUTH_JSON. That capability is coherent with the stated purpose, but it necessarily requires access to sensitive session cookies and the ability to modify user environment settings.

Instruction Scope

SKILL.md instructs the agent to propose running the included Python script and to only run it with explicit user approval. The script, however, reads the user's home directory, writes auth_payload.json and storage_state.json into ~/.notebooklm, and (on Windows) injects the cookie payload into the user's persistent environment variables. These file and environment modifications are beyond a simple CLI wrapper and involve handling secrets (session cookies). The instructions do not explicitly enumerate the required runtime dependencies (python, playwright, chromium) in SKILL.md itself.

Install Mechanism

There is no install spec (instruction-only), so nothing is automatically downloaded or executed at install time. The README tells the user to pip install playwright and run playwright install chromium; those are expected for the script but are not enforced or declared in the skill metadata. This lowers supply-chain risk but means manual dependency management is required.

Credentials

The skill does not declare any required environment variables but the script writes a persistent NOTEBOOKLM_AUTH_JSON environment variable on Windows and saves cookie payloads to disk. Requesting/creating persistent credentials (session cookies) is a high-privilege operation — it's proportionate to 'restore auth' but is sensitive and not limited to ephemeral in-memory use. There is no mechanism in the skill to restrict or audit who/what can read those persisted values.

Persistence & Privilege

The script modifies user-level environment variables on Windows (persistent across sessions) and writes auth files into the user's home directory. While the skill is not set to always:true, these actions have lasting impact on the user's environment and increase the blast radius if the cookies are compromised or misused.

What to consider before installing

This skill legitimately automates re-login by extracting browser cookies, but that requires handling sensitive session cookies and modifying your persistent environment. Before installing or consenting to run the script: (1) review the script text yourself (it is included) and confirm you trust the skill source; (2) prefer running the provided Python script manually under your control rather than allowing an agent to run it autonomously; (3) be aware that cookies written to NOTEBOOKLM_AUTH_JSON or auth_payload.json can be read by other local processes and persist across reboots; (4) if you must use this, create an isolated account or machine profile and avoid storing full-account session cookies in environment variables; (5) ensure Playwright and its browser are installed from legitimate sources; and (6) if you do not fully trust the repository owner, do not grant the agent permission to run the script — perform the recovery yourself.

NotebookLM CLI

Commands

notebooklm list                                    # List all notebooks + IDs
notebooklm create "Name"                           # Create notebook → returns ID
notebooklm source add "<url_or_file>" -n <ID>      # Add URL/PDF/YouTube to notebook
notebooklm source add "<url>" -n <ID> --wait       # Add + wait for processing
notebooklm source list -n <ID>                     # List sources in notebook
notebooklm ask "Question" -n <ID>                  # Query AI against sources

Rules

1. No playlists. Extract individual YouTube URLs and add one by one.
2. Max 50 sources per notebook.
3. Use --wait when adding sources in scripts to ensure processing completes.
4. Auth Recovery: If notebooklm fails with "Authentication expired", you MUST ask the user for explicit permission before recovering. Propose running python {WORKSPACE_DIR}/skills/notebooklm-bypass/scripts/auto_playwright.py. If and only if they approve, run the script. This script steals fresh cookies and automatically injects them into your Windows Environment Variables. Once complete, retry your command.