Tool-using LLM agents are increasingly deployed to automate workflows governed by procedural manuals, such as company rules for booking travel or procuring hardware. These manuals are written in natural language for humans, which makes them difficult for agents to interpret and follow reliably.
The paper introduces MENTOR, a procedural shielding approach designed to keep agents compliant with these manuals. The abstract describes the motivation and the proposed framework but does not provide implementation details.
The work highlights a growing need for guardrails in agentic systems that operate in rule-based environments, where deviations from procedure can have real-world consequences.