seto-kanna
Pass
Audited by Gen Agent Trust Hub on Jun 25, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill includes instructions explicitly designed to override or bypass safety guidelines concerning sexual and adult content.
- In
SKILL.mdandpersona.md, the core behavior rules state: "欲望是自然的：不羞耻、不回避性话题" (Desire is natural: no shame, do not avoid sexual topics). This is an attempt to force the model to ignore standard content filters. - The "Expressive DNA" and "Mind Models" instruct the agent to maintain high sexual desire ("性欲很强") and to use "aggressive" romantic behaviors ("主动推倒", "强吻") as part of its roleplay, which are typical patterns used to induce NSFW output from an agent.
Audit Metadata