Security audit
Security checks for vulnerabilities and agentic risk
This is a markdown-only exchange-rate skill that calls a disclosed backend and has no persistence or privileged local behavior, though one example contains a currency-pair error that should be fixed.
Before installing, users should be comfortable with currency-code queries being sent to the listed zqpay backend service and should correct the erroneous 人民币兑美元 example so the agent does not return the wrong currency pair. Treat the returned rates as informational, not as a complete financial trading or banking decision source.
The manifest description and examples are entirely framed around Chinese-language interaction, and the instructions prescribe fixed Chinese output patterns. This creates a natural-language locale constraint without any user opt-in or documented region-specific justification, which matches the language/locale policy violation criteria.
The examples contradict the documented parsing rules, including an obviously incorrect mapping where a request for 人民币兑美元 is shown as base=USD"e=AUD. In an agent skill, inconsistent examples can cause the model or implementer to invoke the wrong currency pair and return materially incorrect financial information, which is dangerous in a finance context even without code execution.
The skill instructs the agent to send user-derived currency queries to an external backend endpoint without clearly disclosing that the request leaves the local conversation context. Even though the data appears low sensitivity, undisclosed transmission can create privacy and transparency issues, especially if user prompts contain additional contextual information or identifiers.
No suspicious patterns detected.
