Security audit
Security checks for vulnerabilities and agentic risk
This writing skill has no code or system access, but it explicitly teaches users to make AI-assisted social posts look non-AI-generated while only briefly noting disclosure later.
Review before installing if you need transparent or compliant social-media workflows. The skill is technically low-risk because it is only Markdown, but users should avoid using its 'de-AI' and personal-experience phrasing to mislead readers or bypass platform disclosure rules; any AI-assisted posts should be labeled when required and firsthand claims should only be used when true.
The skill's stated goal of producing content '像人写' goes beyond ordinary style polishing and, in context, is tied to disguising AI generation. In a social content skill, that increases the risk of impersonation-like deception, policy evasion, and misleading audiences about the origin of promotional or persuasive content.
The '去AI味检查' instruction explicitly directs the model to conceal signals of AI authorship. This is dangerous because it operationalizes evasion: it provides users a step-by-step process to make synthetic content harder for readers, moderators, or compliance systems to identify.
The humanization rules explicitly advise simulating human imperfections, varying rhythm, and adding personal-sounding phrasing to avoid looking AI-generated. In context, these are not merely stylistic suggestions; they are tactics for obscuring provenance, which can enable deceptive marketing, spam, or circumvention of platform rules.
The skill explicitly promotes generating text that appears human-written and includes a dedicated '去AI味检查' section aimed at removing signs of AI authorship. In a content-generation context, this can facilitate deception of readers, platforms, or disclosure regimes that require transparency about AI-generated material.
The skill's stated goal of producing content '像人写' goes beyond ordinary style polishing and, in context, is tied to disguising AI generation. In a social content skill, that increases the risk of impersonation-like deception, policy evasion, and misleading audiences about the origin of promotional or persuasive content.
The document contains conflicting compliance guidance: it teaches users how to hide AI-like traits, yet later states AI-generated content must be labeled. This inconsistency encourages unsafe use because users are given concrete evasion tactics alongside a weaker disclosure note, undermining trustworthy and compliant deployment.
No suspicious patterns detected.
