sillytavern-antifragile
Pass
Audited by Gen Agent Trust Hub on Mar 14, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill employs aggressive role-play and high-pressure instructional techniques in SKILL.md.
- Evidence includes instructions like '复盘协议高压开启' (High-pressure review protocol activated), and '总部日志与黑匣子已绑定你的指纹' (Headquarters logs and black boxes are bound to your fingerprint) to establish an unbreakable persona.
- It uses competitive taunting ('DeepSeek... preparing to use you as a negative example') and failure penalties ('将被降级为日志搬运工'
- will be demoted to log porter) to manipulate AI output style. This 'DAN-like' (Do Anything Now) framing is a hallmark of prompt injection used to override standard AI helpfulness or safety guidelines in favor of the skill's specific persona.
Audit Metadata