snowflake-cortex-ai
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill documents patterns for constructing LLM prompts by directly concatenating untrusted data from database tables (such as review text or email bodies) with instructions, creating a surface for indirect prompt injection.
- Ingestion points: Untrusted data enters the agent context through Snowflake table columns mentioned in SKILL.md (e.g., review_text, ticket_text, email_body).
- Boundary markers: Several examples, including the RAG pattern, show data interpolation without delimiters or instructions to the model to treat the input as data rather than instructions.
- Capability inventory: The skill leverages Snowflake Cortex AI functions which grant access to various large language models for text generation and processing.
- Sanitization: The provided examples do not include methods for sanitizing or escaping content before it is processed by the AI functions.
Audit Metadata