customer-research

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

• Third-party content exposure detected (high risk: 0.95). The required runtime workflow uses UnifAPI to fetch live public community text (e.g., Reddit comments via seo/serp → reddit/posts/{id}/comments, TikTok comments via tiktok/videos/{id}/comments), which is outsider-authored free-form prose that the agent will ingest into its LLM context for clustering and quote extraction.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly requires runtime calls to the UnifAPI MCP server (e.g., https://mcp.unifapi.com and the unifapi endpoints) to fetch live Reddit/TikTok/News/SEO comment content that is injected verbatim into the agent's context, so remote content from that URL directly controls prompts at runtime.