The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on several shell-based commands for its initial setup and authentication, including pip install, playwright install chromium, and notebooklm login.
[DATA_EXFILTRATION]: The skill requires access to sensitive Google authentication cookies, which are either read from the environment variable NOTEBOOKLM_AUTH_JSON or from a configuration file at ~/.notebooklm/storage_state.json. It also provides methods to read local files via the add_file API, which could be used to access sensitive local data.
[EXTERNAL_DOWNLOADS]: The skill initiates the installation of the notebooklm-py Python package and the Playwright framework, the latter of which downloads and installs Chromium browser binaries to the local environment.
[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it processes untrusted content from the web and local files.
Ingestion points: Untrusted data is ingested through client.sources.add_url, client.sources.add_file, and client.research.start (as documented in SKILL.md and references/workflows.md).
Boundary markers: There are no explicit boundary markers or instructions to ignore embedded commands mentioned in the interaction patterns for the chat.ask functionality.
Capability inventory: The skill has the capability to perform network-based research, read local files, and generate and download various artifact types (e.g., MP3, MP4, PDF, PNG, JSON).
Sanitization: The instructions do not describe any sanitization or validation processes for the ingested content before it is indexed and queried.

notebook-lm