aminer-deep-search

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

• Third-party content exposure detected (high risk: 0.75). Yes: runtime LLM context includes outsider-authored text from public AMiner API responses (e.g., paper titles/abstracts/venues in search_papers/get_reference_papers), which are JSON-stringified in react_agent.py:_tool_result_message and appended to the chat history via ReactPaperCollector.run().