The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted content from Google NotebookLM notebooks, creating a surface for indirect prompt injection where instructions embedded in a notebook could influence agent behavior.
Ingestion points: Notebook content is ingested via the ask_question tool, and library metadata is ingested via list_notebooks (SKILL.md, references/mcp-tool-reference.md).
Boundary markers: The skill includes a 'Verification Protocol' in SKILL.md recommending evidence checks, but the provided scripts (scripts/notebooklm-helper.py) and code examples lack technical delimiters or wrappers to isolate untrusted notebook content from the agent's internal reasoning.
Capability inventory: The skill possesses the capability to modify the user's library through add_notebook, update_notebook, and remove_notebook, which could be exploited if the agent obeys instructions hidden within processed research data.
Sanitization: No explicit sanitization, escaping, or filtering of external notebook content is present in the local helper utility or the provided usage examples.

notebooklm-management