The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted product information retrieved from the external Ozon marketplace via the MPSTATS API. This creates a potential surface for indirect prompt injection where malicious content embedded in marketplace listings could attempt to influence the agent's behavior. * Ingestion points: Data is ingested via scripts/ozon_product_search.py, which fetches product identity cards from the API and saves them to the local filesystem. * Boundary markers: Absent; the skill instructions in SKILL.md do not include explicit delimiters or specific instructions for the agent to treat retrieved product data as untrusted. * Capability inventory: The scripts/ozon_product_search.py script possesses network access (to the NexScope proxy) and filesystem write permissions (to save responses to the nexscope/ directory). * Sanitization: Absent; the skill does not perform sanitization or filtering of the retrieved marketplace content before displaying it to the user or providing it to the agent.

ecommerce-ozon-product-search