cosmosdb-vector-search
Pass
Audited by Gen Agent Trust Hub on Oct 27, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill provides educational content and implementation templates for Azure Cosmos DB vector search. All code snippets follow best practices for security, such as using parameterized queries to prevent SQL injection.
- [INDIRECT_PROMPT_INJECTION]: The skill defines patterns for Retrieval-Augmented Generation (RAG) which involves processing user search queries and external document content. This represents a standard injection surface area for AI agents.
- Ingestion points: External data enters via the
search_queryand retrieved document content processed in thevector_searchmethods across Python, .NET, Java, and JavaScript examples inreferences/vector-repository-pattern.md. - Boundary markers: As these are architectural templates, they do not include specific LLM prompt delimiters or instructions to ignore embedded commands, which should be handled by the consuming application.
- Capability inventory: The skill demonstrates database read/write capabilities (insert, upsert, query, read) using official Azure Cosmos DB SDKs.
- Sanitization: The skill correctly implements SQL parameterization for embedding vectors and query limits to ensure database-level security.
Audit Metadata