dialogue-grounded

PolicyGuard is a sub-agent verifier that enhances LLM agent policy adherence by providing contextual reasoning and conversation-specific feedback across multi-turn interactions, achieving significant improvements on the tau^2-BENCH benchmark.