Categories Companies Research

Loading the SOTA2 catalog…

Explore

Categories Companies Research

Contribute

Join SOTA2 Contact

SOTA2

© 2026 SOTA2

Build in Bellevue, USA

WizWand is now SOTA2 Research

Papers Benchmarks Datasets Tasks

Malicious Skill Detection research benchmarks · SOTA2 Research

Malicious Skill Detection

Benchmarks

Dataset NameSOTA MethodSort

Dataset NameSOTA methodMetricTrendResultsLast Updated

Adversarial Set 400 skills

0Attack Success Rate (ASR)

28

Jun 17, 2026

MaliciousAgentSkills

9B

13

Apr 1, 2026

SKILLVETBENCH labeled set 1.0 (78 malicious + 22 benign skills)

0.95Balance

9

Jun 16, 2026

95Overall Detection Rate

9

Jun 2, 2026

ClawHub Unsafe File Ops 1.0 (n=10)

100Catch Rate

9

Jun 2, 2026

ClawHub Prompt Injection 1.0 (n=19)

100Catch Rate

9

Jun 2, 2026

ClawHub Command Injection 1.0 (n=27)

100Catch Rate

9

Jun 2, 2026

ClawHub Overall 1.0

95Overall Balance

9

Jun 2, 2026

MaliciousAgentSkillsBench (404 malicious, 502 benign)

100Recall

9

May 13, 2026

Clean Base Set 100 skills

99Not Flagged Rate

7

Jun 17, 2026

100-skill dataset (OpenClaw)

SkillScanner (LLM)

98TPR

7

Jun 11, 2026

93True Positives (TP)

6

Mar 31, 2026

906-skill benchmark

Behavioral integrity (BIV)

94.6F1 Score

4

Jul 1, 2026

98.6Ori Score

3

Jul 3, 2026

SkillJect (Malicious)

98Originality

3

Jul 3, 2026

Reviewed Sample (65 Malicious, 13 Benign)

Locate-and-Judge

62True Positives (TP)

3

Jun 23, 2026

SkillFortifyBench (total)

100Precision

1

Mar 4, 2026

—

—Primary metric

0

May 29, 2026

Popular datasets

ClawHub SkillJect SkillFortifyBench OpenClaw MalSkillsBench

Follow for updates

Join SOTA2 Contact