Security audit
Security checks for vulnerabilities and agentic risk
This skill is a simple prompt generator with no executable code, but it forces uploaded-person replacement prompts to use a Chinese character even when the user did not ask for that.
Review before installing if you do not want automatic demographic constraints in image prompts. The skill appears technically low-risk, but users should only use it when they specifically want this replacement behavior and should consider editing the instructions to make identity traits optional and user-directed.
The skill hard-codes a protected ethnicity/nationality attribute ('Character must be Chinese') without user opt-in or task necessity. This creates discriminatory behavior in model outputs and can systematically force sensitive-attribute inference and transformation from uploaded images, which is unsafe and policy-relevant even if it is not a traditional exploit primitive.
Repeating the mandatory requirement that the replacement character 'must be Chinese' reinforces a non-optional sensitive-attribute constraint throughout the skill logic. In the context of image-based person replacement, this increases the chance of biased or exclusionary outputs and normalizes transformation of people into a specific ethnicity regardless of user need or consent.
The listed triggers include generic phrases like "swap person" and "replace model," which can overlap with ordinary image-editing requests and do not clearly distinguish this skill from other editing workflows. The file also does not provide exclusion conditions or negative examples to clarify when the skill should not be used.
No suspicious patterns detected.
