Extracting Threats from System Descriptions with LLMs Comparing One and Two Agents Strategies