notebooklm
Fail
Audited by Gen Agent Trust Hub on Jun 14, 2026
Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill installs the
notebooklm-pypackage and its dependencies (likerookiepy) from PyPI and directly from GitHub repositories. It also triggers the automatic download of the Chromium browser (~170 MB) during its first execution. - [REMOTE_CODE_EXECUTION]: The skill uses
curlto fetch the latest release tag from the GitHub API (api.github.com/repos/teng-lin/notebooklm-py/releases/latest) and dynamically incorporates this untrusted string into apip installcommand, which executes code from the remote repository. - [COMMAND_EXECUTION]: The instructions include a command to install system dependencies using
sudo(sudo playwright install-deps chromium), which represents a privilege escalation risk if executed by an automated agent. - [DATA_EXFILTRATION]: The skill handles sensitive authentication data, specifically Playwright
storage_state.jsonfiles containing Google session cookies. It includes features to extract these cookies from local browsers and allows passing them as inline JSON via theNOTEBOOKLM_AUTH_JSONenvironment variable. - [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as its core functionality involves ingesting untrusted data from URLs and documents (
notebooklm source add). - Ingestion points: Web URLs, YouTube videos, and local files (PDF, PDF, Audio, Video, etc.) added as sources via
notebooklm source add. - Boundary markers: No explicit delimiters or instructions to ignore embedded commands are provided when the agent processes source content.
- Capability inventory: The skill can write files to the local system (
notebooklm download), perform network requests to Google APIs, and execute shell commands via the CLI tool. - Sanitization: There is no evidence of sanitization or filtering of the content retrieved from external sources before it is processed by the agent.
Recommendations
- AI detected serious security threats
Audit Metadata