The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data by enabling users to add various sources such as URLs, YouTube links, local files, and Google Docs into a NotebookLM instance for analysis.
Ingestion points: Untrusted data enters the agent's context through the add_source workflow defined in SKILL.md and routed via scripts/action_router.py.
Boundary markers: There are no specific instructions for the agent to use delimiters or ignore potentially malicious embedded instructions within the ingested source material.
Capability inventory: The skill utilizes powerful browser automation capabilities including navigation, element interaction (clicking/typing), and screen capture across its script suite.
Sanitization: The instructions recommend extracting main content and stripping navigation/ads for synthesized content, which serves as a basic form of sanitization, though it is not applied to all source types.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill performs legitimate network navigation and data transmission to the official Google NotebookLM service.
The skill interacts with notebooklm.google.com to manage notebooks and generate content as part of its primary functionality.

notebooklm