langchain-and-llamaindex-on-azure-sql
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [External Package Dependencies]: The skill recommends several standard Python packages for database connectivity and AI orchestration, such as LangChain, LlamaIndex, and SQLAlchemy. These are well-established libraries in the developer community.
- [SQL Command Execution]: The skill's primary function involves enabling an agent to interact with an Azure SQL database via SQL queries. It correctly identifies that default framework configurations may allow data modification (DML) and provides instructions to implement read-only access at the database level as a security best practice.
- [Data Exposure Considerations]: The documentation notes that default settings in some frameworks may include live database rows in the model's prompt context. It provides configuration steps to mitigate this by disabling row sampling in the schema metadata.
- [Indirect Prompt Injection Surface]: The skill facilitates the processing of external data (database schema and rows), which is a common surface for indirect prompt injection. The author provides clear mitigation strategies:
- Ingestion points: Data enters the context via the database schema and sample row tools described in SKILL.md.
- Boundary markers: The skill suggests using configuration parameters like sample_rows_in_table_info=0 to limit data ingestion.
- Capability inventory: The agent utilizes toolsets that execute SQL commands, creating a pathway for database interaction.
- Sanitization: The skill emphasizes using least-privilege database logins to provide a robust security boundary regardless of the agent's output.
Audit Metadata