Security audit
Security checks for vulnerabilities and agentic risk
This skill is a disclosed business-email drafting aid with some quality and privacy wording caveats, but no evidence of hidden access, exfiltration, or unsafe automatic actions.
Install only if you are comfortable providing business email context to your agent environment. Review generated outreach for accuracy, recipient consent, unsubscribe or opt-out needs, and regional marketing rules; also verify non-English output before sending because some language selections may produce English drafts for later translation.
The trigger phrase at this line is broad enough to activate during ordinary business discussion, not just when the user explicitly wants this email-writing skill. Over-broad activation can cause unintended context loading and unnecessary exposure of user-provided business content to the skill workflow.
This trigger overlaps with common business terminology and may cause the skill to run when users are merely discussing quotations rather than asking to write a quotation email. Such accidental invocation increases the chance of irrelevant prompt injection surface and inadvertent handling of commercial data.
A generic English trigger like this lacks enough specificity for reliable scoping and can match many unrelated conversations. In skill-based systems, weak scoping raises the risk of accidental activation, unnecessary prompt expansion, and avoidable data exposure.
The skill claims that generated email content is processed and retained only locally, but the artifact is an AI skill prompt and does not itself guarantee local-only execution or storage. This can mislead users into providing sensitive customer, pricing, or business data under false privacy assumptions, creating confidentiality and compliance risk.
This markdown file presents the skill entirely in Chinese, starting with the title and section headings, while the policy requires flagging language or locale constraints that force a specific language without user opt-in. There is no note that the templates are intended only for Chinese-speaking users or any option to select another language.
The guide presents region- and language-specific communication rules as prescriptive defaults rather than optional, contextual suggestions. In an email-writing skill, this can cause the agent to apply cultural stereotypes or locale-specific behavior without explicit user consent, leading to biased, inappropriate, or misleading business communications.
The manifest explicitly describes professional email writing in multiple languages, including use cases for multilingual business emails. In code, inquiry replies for Arabic, Spanish, and Russian do not generate those languages at all; they fall back to English and prepend a note saying it should be translated later. The same pattern appears in other generators, so the implemented behavior is narrower than the advertised multilingual capability.
The file-level description claims support for CN/EN/AR/ES/RU, but for unsupported template branches the generator prepends a note and produces English content instead. This effectively forces English output without user opt-in, which is a natural-language locale policy concern because users selecting ar/es/ru do not actually receive content in their chosen language.
The manifest and module docs present this as a generator for distinct foreign-trade email categories, but order_confirmation is handled by gen_inquiry_reply and after_sales is handled by gen_follow_up. That means the produced content semantics do not match the claimed email type, creating a mismatch between declared behavior and actual output.
规则要求识别自然语言层面的语言/地区政策问题。该文件以中文作为默认说明语言，且未声明这是面向特定中文用户群或提供语言选择，可能与要求用户可选择语言/locale 的组织政策不一致。
The template promotes a structured multi-step follow-up sequence, including repeated contact and urgency cues such as special offers and final reminders, without any guidance on consent, unsubscribe options, frequency limits, or applicable anti-spam/privacy rules. In an email-writing skill for B2B outreach, this can encourage users to generate borderline-harassing or non-compliant campaigns that create legal, reputational, or trust risks.
No suspicious patterns detected.
