Security audit
Security checks for vulnerabilities and agentic risk
This is a narrow teaching skill that gives counterexamples to claims, with no code execution, data access, persistence, or hidden high-impact behavior.
Install this if you want a Chinese-oriented skill for testing claims by looking for exceptions. Be aware it may answer narrowly with counterexamples for broad questions like whether a statement is reliable, so you may need to ask explicitly if you want a full explanation or another language.
The trigger phrases are broad, generic conversational prompts such as asking for exceptions, boundaries, or whether a claim is reliable. In an agentic environment, this can cause accidental or overly eager invocation on ordinary user questions, leading the skill to override more appropriate skills and produce narrowed 'counterexample-only' outputs that may distort user intent.
The instructional content from the title through the output template and examples is written entirely in Chinese and directs the skill's behavior in Chinese, while not giving the user any language or locale choice. Although bilingual metadata exists, the operative instructions do not state that responses should follow the user's preferred language, which creates a locale policy risk.
All user-facing instructions and examples are in Chinese, and the behavior is described without any opt-in or language-selection mechanism. This can amount to forcing a specific language/locale by default, which the policy says to flag unless choice or justification is provided.
No suspicious patterns detected.
