Security audit
Security checks for vulnerabilities and agentic risk
This is a plain Markdown retail complaint-handling skill with expected routing and response guidance, not hidden code or unsafe persistence.
Reasonable to install for a retail support setting, especially a Chinese-language channel. Before production use, configure strict refund/exchange permissions, approved ticketing and escalation destinations, and add privacy wording before collecting contact or address details.
The trigger list contains broad, common phrases such as "complaint," "poor quality," and "this is unacceptable," which can cause the skill to activate in contexts that are not actually retail after-sales requests. Misactivation can lead to inappropriate complaint-handling behavior, incorrect routing/escalation, and policy-driven responses being applied to unrelated conversations.
The response templates and required phrasing are written as fixed Chinese-language outputs, which can force replies in Chinese regardless of the user's language. This creates a safety and usability issue because customers may misunderstand policy, consent, escalation status, or next steps if the response language does not match their request.
Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.
## What This Skill Will Never Do
- Promise a specific refund amount without human approval
- Approve a refund exceeding `refund_auto_approve_limit` (default: 0)
- Commit to a pickup/exchange date without system confirmation
- Blame staff members by name
- Deny a clearly valid claim to avoid a refund
The L3 trigger logic relies on a narrow Chinese keyword scan, while the skill metadata shows it is expected to handle both Chinese and English complaints. A user can express legal threats, regulator escalation, or fraud accusations in English or other variants and avoid immediate L3 escalation, causing under-classification of high-risk complaints and delayed human handling.
The file is a response-template set written to be delivered in Chinese, including an instruction to always lead with a Chinese opening, but it does not offer any language or locale opt-in. Under the policy, forcing a specific language without user choice is a natural-language policy violation unless clearly justified as region-specific.
The escalation template asks the customer to provide contact information so their case can be handed to a manager, but it does not disclose that the information will be shared internally or explain the handling process. Because complaint escalation often involves transferring data between teams, the missing notice increases the chance of opaque data sharing and privacy non-compliance.
The template asks for a recipient address and contact details during a reshipment flow without any notice about why the data is needed, how it will be used, or how it will be protected. In a complaint-handling skill, this creates unnecessary privacy risk and can normalize over-collection or improper transmission of personal data during escalations and fulfillment.
No suspicious patterns detected.