Security audit
Security checks for vulnerabilities and agentic risk
This skill is a Markdown-only Dianping decision aid, with no hidden execution, persistence, or data-exfiltration behavior found.
Installers should understand this skill is tailored to Dianping and may answer in Chinese-style templates. Use logged-in Dianping pages only when you intentionally need coupon or order details reviewed, and avoid sharing unrelated account or payment information.
The trigger examples include generic requests like '找一家适合约会的餐厅' and '推荐个适合久坐办公的咖啡馆' without requiring explicit Dianping intent. This can cause the skill to be selected for broad restaurant or venue recommendation queries, increasing unintended activation and scope creep into general advice workflows the skill was not narrowly declared for.
The file defines outcome labels entirely in Chinese and provides a Chinese-only output pattern, which can impose a specific language on responses. There is no indication that the user can choose another language or that the locale restriction is documented as an opt-in or region-specific requirement.
This markdown file contains natural-language guidance that switches into Chinese for core section headings and later provides the sample comparison output only in Chinese. Because the skill does not state that it is intentionally China/Chinese-language specific or offer a user language choice, it may violate language/locale policy by implicitly forcing a specific language.
The markdown includes only a Chinese-language 'Good output example,' which implicitly steers the skill toward responding in Chinese. Because there is no surrounding instruction offering a language choice or explaining that the skill is specifically limited to Chinese output, this conflicts with the language/locale policy requirement.
The file prescribes a fixed Chinese-language output pattern without considering the user's requested language or locale preferences. This can cause the agent to ignore user intent, reduce accessibility, and create prompt-quality issues where the assistant appears non-compliant or misleading about its capabilities, though it is not a direct code-execution or data-exfiltration risk.
No suspicious patterns detected.
