ToolSafe: Enhancing Tool Invocation Safety of LLM-based agents via Proactive Step-level Guardrail and Feedback Paper • 2601.10156 • Published 16 days ago • 26