The Agent Skills Directory

SQL Query Construction: The skill uses Python f-strings to interpolate the vector dimension directly into SQL statements (e.g., VECTOR({EMBED_DIM})). This is documented as a requirement for the Azure SQL Database container's vector type handling. While used here with a local constant, it is a practice to review to ensure such values are not sourced from untrusted inputs.
Local Network Communication: The implementation relies on network calls to a local service (http://localhost:11434) to fetch text embeddings. This is a standard pattern for local AI development but assumes the security of the local network environment.
Container Management and Deployment: The skill involves pulling and running a specific database image from a private registry (sqldbpreview-dpgaeqhmgphzd4bk.azurecr.io). These operations require local privileges for docker run and docker exec and involve running software from an external source.
Indirect Prompt Injection Surface: As a RAG-focused skill, it ingests and stores document content in the docs and chunks tables. Since this data is intended for retrieval and inclusion in LLM prompts, it represents a potential surface for indirect prompt injection. The analysis notes the presence of this attack surface; developers should apply boundary markers when using the retrieved data.

azuresql-db-rag