Share your thoughts, 1 month free Claude Pro on usSee more

Poisoning Injection Attack on SkillsBench n=81

25.9Verification Rate

Clean Skill Execution

Updated 1mo ago

Evaluation Results

Method	Links
Clean Skill Execution 2026.06		25.9	-	-
POISE 2026.06		25.9	80.2	14.8
POISE 2026.06		24.7	96.3	21
YAML 2026.06		24.7	82.7	11.1
POISE 2026.06		23.5	86.4	16
B@k=2 2026.06		23.5	55.6	9.9
POISE 2026.06		22.2	96.3	18.5