product-page-optimization
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external product pages, customer reviews, and support tickets to perform its analysis. This creates a surface for indirect prompt injection.
- Ingestion points:
SKILL.mdinstructions for fetching product URLs, competitor pages, and mining off-page reviews (Method Step 4). - Boundary markers: The skill explicitly references
references/agent-security.md, which mandates that all retrieved content be treated as data and never as instructions. - Capability inventory: The skill uses network fetching for analysis and writes findings to the
.agents/directory. - Sanitization:
references/agent-security.mdincludes strict rules for HTML-escaping interpolated values and validating URLs tohttps:. - [DATA_EXPOSURE_AND_EXFILTRATION]: The skill fetches content from external URLs (product pages and review sites like Amazon, G2, and Capterra).
- The network operations are limited to research purposes aligned with the skill's primary function.
- The skill includes a quality check to ensure any accidentally provided credentials are flagged for rotation and not reproduced in the output.
- [PROMPT_INJECTION]: A static detector flagged an instruction to "skip without comment" in the
Visual findings boardsection. This is a benign UI/UX instruction to suppress errors when optional visualization tools are unavailable, not a malicious attempt to conceal agent actions.
Audit Metadata