Security audit
Security checks for vulnerabilities and agentic risk
This is a frontend CSS architecture guidance skill with no hidden execution or sensitive access, though some examples are broad or opinionated.
Review the generated UI text and theme requirements before adopting them, especially if your product is not Chinese-language or does not need a light/dark/system toggle. The skill otherwise appears safe as frontend architecture reference material.
The README lists invocation text as "set up the CSS architecture for this project" / "build a design token system" without defining scope limits, exclusions, or when the skill should not activate. These phrases are plausible everyday development requests and the file provides no negative examples or narrower context to prevent unintended invocation overlap with other frontend or design skills.
This markdown file includes user-facing natural-language strings such as the aria-label and button labels entirely in Chinese. Because the file does not indicate that the skill is region-specific or provide an opt-in language choice, it appears to force a specific language for users, which matches the language/locale policy violation criteria.
The manifest says the skill should be used for broad requests involving topics like CSS architecture, layout systems, responsive strategy, and technical handoff specs. While domain-related, it does not provide boundaries or negative examples, so the activation scope may overlap with many general frontend or design discussions and cause unintended invocation.
The statement 'Every new project must include light/dark/system theme toggle' imposes a universal preference-related requirement rather than offering it as a choice based on user or project needs. This can be interpreted as forcing a preference setting approach without opt-in or justification.
The example markup uses Chinese text labels such as "已上线", "主操作", and "次操作" in a general component-guidelines document. Because the file does not state that the skill is China-specific or provide user opt-in for locale, this can be read as forcing a specific language in example outputs.
No suspicious patterns detected.
