Security audit
Security checks for vulnerabilities and agentic risk
This is a markdown-only foreign-trade writing and analysis skill with disclosed business-assistant behavior and no code, persistence, credential access, or hidden execution.
Installers should understand that this skill can produce customer due-diligence, credit, and risk-style business assessments. Use those outputs only as drafting or research assistance, verify facts from authoritative sources, and avoid making adverse legal, compliance, or financial decisions from the model output alone.
The README advertises customer background checks, risk assessment, and credit-rating style analysis without any warning about accuracy limits, lawful basis, privacy constraints, or prohibited uses. In a trade-assistant context, this can encourage users to make consequential business decisions or process third-party data without informed consent, compliance checks, or human verification.
The trigger list contains very broad business terms such as '外贸', '询盘', '获客', and '外贸文案', which can match ordinary user conversation and cause the skill to activate outside clearly intended contexts. Over-broad activation increases the chance of unintended execution on unrelated business chats, potentially exposing users to unrequested profiling, generated outreach content, or background-check workflows.
The instruction says the assistant '帮助中国出口商撰写专业的商务文案', which hard-codes a specific national business audience without indicating user choice or opt-in. This creates a natural-language policy concern because the skill fixes its locale/market orientation instead of explicitly allowing other user contexts.
The trigger section uses broad natural-language cues like writing emails, analyzing replies, follow-up strategy, and customer due diligence without defining clear activation boundaries, exclusions, or confirmation steps. In an agent setting, this can cause unintended invocation on ambiguous user input, leading the assistant to perform sensitive business-analysis or outreach tasks when the user did not explicitly request them.
No suspicious patterns detected.
