agentic-research-workflow

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

• Third-party content exposure detected (high risk: 0.95). The workflow’s research step calls external tools (Tavily web search, Wikipedia summary, arXiv summaries) and then injects their returned text into the LLM context via messages.append({"role":"tool","content": str(result)}), where result is outsider-authored web content fetched at runtime.