Reflect-Guard: Enhancing LLM Safeguards against Adversarial Prompts via Logical Self-Reflection | ArxivCSExplorer