commerce-ad-claim-compliance-kr

Pass

Audited by Gen Agent Trust Hub on Aug 19, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-provided advertising copy and landing page text. This represents a potential surface for indirect prompt injection where malicious instructions could be embedded in the analyzed data to influence the agent's behavior. However, the skill implements a rigid rule-based workflow and a structured output format (tables and specific status codes like BLOCK/WARN/PASS) that effectively constrains the agent's interpretation and output, significantly mitigating the risk.
  • Ingestion points: The 'Advertising Copy / Detail Page Full Text' slot defined in the '워크플로우' (Workflow) section of SKILL.md.
  • Boundary markers: The skill uses a multi-step inspection sequence ('Step 1' to 'Step 5') and forces a specific '위반 표' (Violation Table) output format to maintain task focus.
  • Capability inventory: The skill uses text generation for compliance reporting and mentions invoking external tools (e.g., moai-lawyer:legal-mfds-safety) for regulatory data lookup.
  • Sanitization: The instructions emphasize 'Rules-based verification' and provide specific examples of allowed versus prohibited phrasing, serving as a functional filter for external content.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 19, 2026, 01:41 PM
Security Audit — agent-trust-hub — commerce-ad-claim-compliance-kr