The Agent Skills Directory

[Local Development Credentials]: The skill instructions and code snippets use a placeholder password (YourStr0ng_Passw0rd) for the Docker container environment and connection strings. This is a standard practice for local development sandboxes, but users should follow best practices by rotating these to secure secrets managed via environment variables in more permanent environments.
[Vendor Resource Downloads]: The workflow requires pulling a specialized container image from a private Azure Container Registry (azurecr.io). This registry belongs to the vendor's infrastructure and is used to provide the preview version of the database engine required for vector operations.
[SQL String Interpolation]: The Python code uses f-strings to interpolate the EMBED_DIM variable directly into SQL queries (e.g., CAST(... AS VECTOR({EMBED_DIM}))). The skill identifies this as a technical requirement of the Azure SQL vector preview, which requires the dimension to be a literal integer rather than a bind parameter. In the provided context, this is a fixed configuration value rather than user-supplied input.
[Indirect Prompt Injection Surface]: The skill facilitates the creation of a Retrieval-Augmented Generation (RAG) system, where external documents are stored and later retrieved to provide context for an LLM. This architectural pattern represents an ingestion surface for untrusted data. When implementing the final prompt construction, users should ensure the retrieved content is clearly delimited from agent instructions to mitigate the risk of indirect prompt injection.

azuresql-db-rag