patient-question-content

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

• Third-party content exposure detected (high risk: 0.85). At runtime the workflow calls UnifAPI to fetch public web/search/community/news content (e.g., seo/keywords/*, geo/serp, and reddit/posts/{id}/comments), which includes outsider-authored free-text questions/comments that are ingested into the agent’s LLM context.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly requires connecting at runtime to UnifAPI (e.g., via https://mcp.unifapi.com and related unifapi endpoints) to fetch live search/Reddit/news data which is injected into the agent context and directly controls the agent's prompt-driven outputs, so this is a required runtime external dependency that controls prompts.