Security audit
Security checks for vulnerabilities and agentic risk
This is a Chinese-language critique skill that gives sharp devil's-advocate feedback, with no hidden code execution, data access, or persistence beyond normal skill installation.
Install this if you want a Chinese, blunt devil's-advocate reviewer for ideas. Avoid using it when you want neutral brainstorming, emotional support, or professional advice; in legal, medical, financial, or other high-stakes contexts, treat its output as critique prompts, not final guidance.
The skill description defines activation in very broad natural-language terms such as generic requests to critique, pressure-test, or find flaws in an idea, with no clear exclusions. This can cause the skill to trigger on common brainstorming or advisory prompts outside its intended niche, leading to unintended persona injection and unreliable routing behavior.
The activation section relies on inferred user intent, including vague formulations like asking whether an idea is '靠谱' or '会不会有坑', and explicitly says it applies across all domains. Without constraints or counterexamples, the agent may over-apply this hostile critique mode to ambiguous requests, producing off-target or excessively aggressive responses in contexts where the user did not actually consent to that style.
The entire skill reference is written in Chinese and includes no indication that users may choose another language or locale. Under the stated policy, a forced language choice without user opt-in can be a natural-language policy violation unless the locale restriction is clearly justified.
No suspicious patterns detected.
