ResearchDatasetsMaliciousAgentSkillsBenchFollowBenchmarksTask NameDataset NameSOTA MethodSortMost resultsRecently updatedMost papersApplyTask NameDataset NameSOTA ResultTrendResultsMalicious skill detectionMaliciousAgentSkillsBench (404 malicious, 502 benign)100Recall9Malicious Instruction DetectionMaliciousAgentSkillsBench traditional IPI baselines63.93Precision4