survey

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.80). The skill explicitly instructs runtime fetching/ingestion of external content (e.g., source_add(..., url="https://arxiv.org/pdf/...") and relies on the NotebookLM MCP CLI hosted at https://github.com/jacob-bd/notebooklm-mcp-cli to run queries/ingestions, so external URLs are used at runtime to inject remote content into the model context and thereby directly influence agent outputs.