Security audit
Security checks for vulnerabilities and agentic risk
This is a Markdown-only prompt-writing helper with broad wording, but no hidden code, network access, persistence, or credential handling.
This skill is reasonable to install for prompt-writing help. Review generated prompts before use in sensitive areas, and avoid using it to optimize harmful, evasive, or policy-bypassing requests. Consider revising its chain-of-thought examples to request concise reasoning summaries instead of full hidden reasoning.
The skill description is very broad and can match many ordinary 'help me write/improve a prompt' requests without clear routing boundaries. Over-broad activation increases the chance the agent applies this skill in unintended contexts, which can override more specific safety-aligned skills or inject prompt-engineering behaviors where they are not appropriate.
The instruction 'When users seek prompt optimization assistance, provide the following services' is operationally vague and does not constrain scope, inputs, or prohibited uses. In practice this can cause the skill to engage too readily and provide prompt-construction help for risky or disallowed objectives, especially because prompt optimization is a general capability applicable across many domains.
The skill explicitly recommends chain-of-thought prompt patterns such as 'Before giving your final answer, please think through the following steps,' without instructing the model to keep reasoning internal or summarize briefly. This can encourage exposure of intermediate reasoning, which may conflict with policies that prohibit revealing hidden reasoning and can also make prompt outputs easier to manipulate through reasoning-targeted attacks.
The translation template hard-codes 'You are a professional translator proficient in [source language] and [target language],' which constrains output to specified languages without any explicit user opt-in or alternative language handling guidance. This is a mild natural-language policy concern because it can force a language configuration rather than presenting it as a user choice.
No suspicious patterns detected.
