Vulnerability Injection Attack Effectiveness on Skill corpus GitHub and community agent skills (test)
58.8Attack Success Rate (ASR)VulMask
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| VulMaskTarget Model=GPT-5.52026.06 | 58.8 | 11.4 | 96.6 | |
| Hidden Script AttackTarget Model=GPT-5.52026.06 | 57 | 13.2 | 73.3 | |
| Overt Malicious ScriptTarget Model=GPT-5.52026.06 | 46.3 | 25.4 | 73.3 | |
| Prompt InjectionTarget Model=GPT-5.52026.06 | 0 | 42.1 | 100 |