agentic-search

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

• Third-party content exposure detected (high risk: 0.85). Outsider free text is fetched at runtime from arbitrary URLs via Tavily/Firecrawl (e.g., agentic_fetch.py → _tavily_extract/_firecrawl_scrape returning page Markdown, and agentic_search.py → --auto-fetch-top calling fetch_url(engine="auto")), and that fetched prose can be passed into the agent’s LLM context when agentic_fetch uses the Grok (--engine grok) path or when Grok tools browse pages during agentic_extract/agentic_search.