omnisearch

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

• Third-party content exposure detected (high risk: 0.90). The skill explicitly fetches and ingests untrusted public web and social-media content (e.g., brave_search.py calling Brave /web/search and its LLM Context API, tavily_search.py calling Tavily /search and /extract, xpoz_search.py using the Xpoz SDK for Twitter/Reddit/Instagram, and Firecrawl/Exa cross-skill scrapers) and omnisearch.py consumes those results (including AI-generated summaries/answers) as part of routing and answer-generation, so third‑party content can materially influence agent outputs and actions.