notebooklm

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.90). SKILL.md explicitly ingests arbitrary public web content and user-generated sources via commands like "nlm source add <notebook_id> --url 'https://...'", "nlm source add <notebook_id> --youtube '...'", and the research/import flow ("nlm research start" / "nlm research import <task_id>"), which the agent is expected to read/query and which directly drive generation and actions (queries, audio/video, reports), enabling third-party content to influence behavior.