Security audit
Security checks for vulnerabilities and agentic risk
This is a markdown-only banking regulatory reporting assistant with no executable code or hidden persistence, but users should treat its high-stakes compliance and credit-policy outputs as drafts requiring authorized review.
Install only for bank regulatory reporting or compliance teams that can validate China-specific regulatory content. Do not use its report drafts, credit-policy rules, negative lists, or remediation suggestions as final decisions without review by authorized compliance, risk, legal, and business owners, and avoid entering sensitive customer data unless your environment is approved for it.
The trigger phrases are broad terms such as regulatory reporting and compliance reporting that can match ordinary banking conversations. Over-broad activation increases the chance that the skill is invoked in unintended contexts, exposing users to specialized regulatory guidance or sensitive workflow behavior when they did not request it.
Natural-language policy violations apply to all file types. The skill presents itself bilingually in places, but the operational content, triggers, and example prompts are overwhelmingly Chinese-specific and do not state whether users may choose another language. This can amount to a forced language/locale assumption without explicit opt-in or justification.
The skill is presented as a regulatory reporting assistant, but Appendix G materially expands it into credit policy analysis, rule generation, and compliance assessment workflows. This capability drift can cause the agent to produce operational risk decisions outside its declared scope and governance controls, increasing the chance of unauthorized or under-reviewed high-stakes guidance.
Generating lending rules, industry risk rules, negative lists, and business adjustment guidance goes beyond clerical reporting support into prescriptive risk governance. In a banking context, such outputs can directly influence credit decisions and risk appetite without appropriate human authorization, creating compliance, fairness, and control risks.
The document states that the skill only provides analysis suggestions and does not replace decision-making, yet adjacent content instructs it to generate actionable rules, negative lists, and pause/exit strategies. This contradiction can mislead users and reviewers about the real authority of the tool, weakening reliance on guardrails and increasing the risk of unsafe operational use.
No suspicious patterns detected.
