Security audit
Security checks for vulnerabilities and agentic risk
This skill only rewrites vague requests into structured prompts and does not include code, tool use, persistence, or data access.
Installers should understand that this skill may turn vague requests into refined prompts instead of doing the requested work immediately. It appears safe from a security perspective, but users may want clearer activation rules and safer wording around reasoning disclosure before relying on it broadly.
The activation criteria are broad enough that the skill may trigger on many normal user requests that are merely concise, causing unsolicited prompt-rewriting instead of responding directly. This creates instruction-routing risk and can degrade user intent fidelity, especially when the agent diverts from the requested task into meta-prompt generation without a clear threshold or opt-in.
The instruction to 'immediately' generate a prompt for vague requests is ambiguous and encourages automatic takeover before evaluating whether direct assistance is more appropriate. In agent settings, this can lead to over-triggering and unsafe or unwanted behavior substitution, where the system changes mode without sufficient justification.
The skill explicitly recommends prompts that ask another AI to reveal its chain-of-thought or internal reasoning process. Requests for hidden reasoning can conflict with model safety boundaries and may be used to coax disclosure of internal deliberations rather than obtaining task-relevant conclusions or concise rationale.
Across the file, the skill's instructions, examples, and operative guidance are presented entirely in Chinese, and the document does not tell the assistant to preserve or ask for the user's preferred language. For a general-purpose skill, this can create an implicit language policy constraint without user opt-in.
This JSON eval file contains natural-language prompts in Chinese and mixed Chinese-English, but nowhere documents whether the skill should preserve the user's language, ask for language preference, or support multilingual output. Because SQP-3 applies to all file types, this can indicate an implicit language/locale behavior without user opt-in.
No suspicious patterns detected.
