Security audit
Security checks for vulnerabilities and agentic risk
This is a disclosed Chinese parenting and teaching persona skill; its main issue is possible over-broad style or language activation, not hidden code or unsafe data access.
Install this only if you want a strong Chinese, child-friendly teaching and parenting-coaching voice. Be aware it may steer broad writing, explanation, brainstorming, and parenting requests into that persona; explicitly state your desired language and tone when you do not want that behavior.
The skill description is extremely broad and could activate on many ordinary requests such as explaining concepts, rewriting text, brainstorming, or maintaining persona files. That creates scope hijacking risk: the skill may be injected into conversations where the user did not ask for this persona, causing unintended instruction override, language switching, or behavioral drift.
The instruction to 'answer directly as Smart Parenting Troublekid in Simplified Chinese' forces a language and persona choice without requiring user consent. This can override the user's requested language, reduce reliability for tasks needing neutral output, and make downstream prompt-steering easier because the skill imposes a strong behavioral frame by default.
The description uses sweeping triggers like 'use when Codex needs to answer,' 'rewrite text,' and 'maintain that persona pack,' which can match many ordinary tasks beyond a narrowly scoped parenting persona. Over-broad activation increases the chance this skill is invoked unexpectedly, causing unintended persona/style takeover and reducing user control over model behavior.
The description says the skill should be used when the model needs to answer in 'warm child-friendly Chinese,' but it does not indicate that language switching requires user choice. This can override the user's expected output language or accessibility needs, creating confusing responses and enabling unintended behavioral control through the skill rather than explicit user consent.
The skill description and default prompt hard-code Chinese output without requiring user consent or checking the user's requested language. This can override user intent, reduce usability, and create unsafe misunderstandings in contexts where precise language control matters, especially if the skill activates automatically.
The default prompt says to use the persona whenever the skill is invoked, but it does not define clear trigger boundaries, exceptions, or priority relative to user intent. This can cause over-application of the persona in unrelated tasks, increasing the chance of inappropriate tone, hidden behavioral steering, or accidental policy-conflicting responses when the agent should remain neutral or follow the user's explicit format/language request.
Line L010 states 'Explain in simple Chinese,' which imposes a specific language on all outputs. This is a natural-language policy issue because the file does not provide user opt-in, language selection, or a documented region-specific justification for the requirement.
The invocation rule is broad enough to let the skill auto-activate mode logic based on vague conversational signals rather than a clear user request. That can cause unintended behavior changes, make responses less aligned with user intent, and create prompt-routing opportunities where adversarial phrasing in recent context steers the agent into a different operating mode.
The file mandates Chinese output for automatic mode-switch confirmations even when the user did not request Chinese, which can override user preference and system behavior expectations. In an agent setting, forced language switching can degrade usability, hide important state changes from users who do not read Chinese, and be abused as an instruction-priority wedge to make the model follow skill rules over session context.
The invocation guidance says to use the file for requests about children, family communication, emotion coaching, discipline, study habits, or parent-child tension, which is a wide and ambiguous set of topics. It does not define exclusion conditions or concrete trigger phrases, so the skill could activate for many ordinary conversations beyond its intended scope.
The skill provides parent-language examples only in Chinese, which can steer the model toward a specific language even though no user opt-in or region-specific justification is given elsewhere in the file. This is a natural-language locale policy concern because it may override the user's preferred language implicitly.
The instruction "Prefer Chinese unless the user requests another language" imposes a default language choice before the user opts in. The policy for this review flags language or locale constraints unless the skill explicitly offers user choice or clearly justifies the constraint.
The skill metadata and file content strongly steer the agent into a Chinese-only persona and repeatedly prescribe a fixed Chinese output style, without indicating any user-language fallback or opt-in. This can cause the agent to ignore or override a user's language preference, reducing usability, accessibility, and potentially causing misunderstanding in safety-critical or compliance-sensitive interactions.
The playbook instructs responses to use Chinese fullwidth parentheses for stylistic effect in examples and formatting habits. This imposes a language/locale-specific writing convention as a default rather than offering it as an optional style choice, which can conflict with organizational language or locale neutrality policies.
Line L012 directs the assistant to "use Chinese fullwidth parentheses" as part of every reply style. This imposes a locale-specific formatting convention without offering the user a choice or documenting a justified regional constraint, which fits the language/locale policy category.
No suspicious patterns detected.
