T09 · Insecure Skill Coding Practices
Warning
- Location
- SKILL.md:59
- Finding
- Generation of Unverified Warranty, Safety Certification, and Customer Support Claims<![CDATA[ ## Vulnerability Details **File Location**: `SKILL.md`, line 59 **Vulnerability Type**: Unverified commercial and product-safety claims **Risk Level**: Medium ### Complete Code Snippet ```markdown - Adds trust signals Western buyers expect (warranty language, safety certs, customer support) ``` ### Technical Analysis The skill instructs the agent to add warranty language, safety certifications, and customer-support statements as buyer trust signals. It does not require these claims to be present in the source material or verified by the user before publication. Although the skill later calls for Amazon policy-compliance validation, it defines no control that distinguishes verified product attributes from model-generated marketing content. Consequently, the agent may fabricate or overstate warranties, certifications, testing results, or support arrangements while following the documented workflow. This is an insecure skill instruction because safety- and compliance-relevant claims must be derived exclusively from authoritative product records. The finding does not provide access to system privileges, files, credentials, or code execution; its scope is the integrity and compliance of generated listing content. ### Attack Path 1. A user supplies product information that contains no warranty, certification, or customer-support terms. 2. The skill parses the supplied information and applies its cultural-adaptation instructions. 3. To satisfy line 59, the agent introduces a warranty, safety certification, or support claim that was not provided or verified. 4. The generated content is copied into an Amazon product listing without independent review. 5. Customers or marketplace reviewers rely on the unsupported claim, potentially resulting in misleading advertising, product-safety misconceptions, marketplace enforcement, refunds, or legal exposure. ### Impact Assessment No operating-system privileges or technical access can be obtained through thi ...[truncated 493 chars]
- Remediation
- <![CDATA[ ## Remediation Suggestions Replace the instruction with a strict evidence-based rule: ```markdown - Include warranties, safety certifications, testing claims, and customer-support terms only when explicitly provided and verified by the user. Never infer or invent these claims. If relevant information is missing, omit the claim or mark it as requiring seller confirmation. ``` Additional hardening measures should include: 1. Separate user-provided facts from model-generated marketing language during input parsing. 2. Maintain an allowlist of claims that may be reproduced only when supported by source data. 3. Require the final validation step to flag every warranty, certification, safety, testing, and support statement with its source. 4. Use explicit placeholders such as `[CERTIFICATION REQUIRES VERIFICATION]` rather than generating plausible certification names or warranty periods. 5. Warn the user that generated listings require legal, regulatory, and marketplace-policy review before publication. 6. Reject conflicting or ambiguous certification information instead of resolving it through inference. ]]>