ToolSafe: Enhancing Tool Invocation Safety of LLM-based agents via Proactive Step-level Guardrail and Feedback Paper • 2601.10156 • Published Jan 15 • 26
