cardiology-youtube-scriptwriter

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

• Third-party content exposure detected (high risk: 0.90). The skill's Phase 2 "Social Listening Research" and social-listening.md explicitly instruct the agent to perform web searches and mine public, user-generated sources (e.g., site:reddit.com queries, YouTube comment mining, public forums and news) and to extract and synthesize those findings as part of its workflow, exposing it to untrusted third-party content that could carry indirect prompt injection.