manage-product-catalog
Pass
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and manage external product data, including AI-generated descriptions and content from external marketplaces (Shopee, Lazada, TikTok Shop). This creates a surface for indirect prompt injection where malicious instructions could be embedded in product metadata to influence the agent's behavior during catalog processing.
- Ingestion points: Product creation/update flows, marketplace synchronization scripts, and ERP/PIM data pipelines (referenced in
SKILL.mdunder Multi-channel Sync). - Boundary markers: The skill partially mitigates this by requiring a 'mandatory human-review gate' for AI content and using Pydantic schema enforcement to reject 'schema-drifted entries'.
- Capability inventory: The skill utilizes
write_file,edit_file, andfetchto update the catalog and interact with external APIs. - Sanitization: While schema validation is present, explicit sanitization or escaping of natural language product descriptions before they are processed by LLM-based tools (like the GraphRAG or semantic search layers) is not explicitly detailed.
Audit Metadata