product-catalog-audit
Pass
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external, untrusted content including product descriptions from CSV/catalog exports and content fetched from live storefront URLs. This creates a surface for indirect prompt injection where malicious instructions embedded in a product description or webpage could attempt to steer the agent's behavior.
- Ingestion points: Catalog exports (provided via CSV, paste, or MCP) and storefront web pages fetched during price/availability verification.
- Boundary markers: The skill explicitly references
references/agent-security.md, which contains instructions to treat every retrieved byte as data rather than instructions, and to report injection attempts rather than following them. - Capability inventory: The skill can fetch external URLs and read/write to the
.agents/directory (e.g.,product-context.mdandstore-loop-ledger.md). - Sanitization: Instructions include HTML-escaping interpolated values and quoting reported content to prevent propagation of instructions.
Audit Metadata