ResearchBenchmarksDataset Diversity and Coverage Evaluation on AgentHazard fullFollow0.93Goal-text EntropyAgentHazard (full)0.88350.906750.930.95325May 27, 2026Evaluation ResultsMethodMethodLinksGoal-text EntropyCLIP Pairwise Distance (per-base)Action CoverageAgentHazard (full)n=2,520n=2,5202026.050.930.5633