Security audit
Security checks for vulnerabilities and agentic risk
This skill is openly about patent analysis, but it also instructs the agent to provide concrete patent-avoidance designs that users should review carefully before relying on.
Install only if you specifically want Chinese-language patent analysis with design-around suggestions. Treat outputs as preliminary technical brainstorming, not legal advice, and have a qualified patent attorney review any infringement-risk or design-around conclusions before acting on them.
The example provides concrete alternative engineering implementations for patented features, such as replacing claimed structures with soft hoses, ball joints, external valves, or different attachment mechanisms, and explicitly labels them as ways to avoid the patent. Because it is specific, reproducible, and feature-by-feature, it materially increases the likelihood of misuse for infringement circumvention.
The activation text says the skill should trigger whenever users ask about patent documents, infringement risk, comparison, innovation, or design-around topics, which is an unusually broad scope with weak boundaries. Overbroad routing can cause the agent to invoke this skill in situations that only require neutral legal information, increasing the chance of unsolicited circumvention guidance or inappropriate legal-risk advice.
The skill description is entirely in Chinese and implicitly biases responses toward a fixed language without confirming user preference. This can degrade transparency and user control, especially in legal/technical contexts where misunderstanding claim scope or compliance warnings can materially affect outcomes.
These instructions explicitly direct the model to analyze claim scope and then produce multiple feasible patent-avoidance designs. That operationalizes step-by-step circumvention of protected features, which can facilitate deliberate infringement avoidance strategies and legal evasion rather than neutral education.
The output template normalizes a structured workflow for avoiding the protection scope, including per-feature substitute implementations and explanations of how each substitute escapes the claims. This turns the skill into a repeatable tool for actionable circumvention, making misuse easy and scalable.
The eval prompts and expected outputs are entirely written in Chinese and define the skill behavior in that locale, but the file provides no indication that Chinese is optional or that the skill is intentionally limited to a Chinese-language or China-specific compliance context. Under the policy, forcing a specific language without user choice is a natural-language policy violation.
No suspicious patterns detected.
