genai-agents-setup
Pass
Audited by Gen Agent Trust Hub on Mar 8, 2026
Risk Level: SAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill provides documentation and templates for utilizing the Databricks CLI via subprocess calls to perform legitimate configuration updates on Genie Spaces.
- [EXTERNAL_DOWNLOADS]: The skill references the official databricks-solutions/ai-dev-kit GitHub repository as an upstream source for synchronization and versioning, which is a trusted organizational source.
- [DATA_EXFILTRATION]: The templates implement secure data handling using MLflow tracing and the Databricks SDK. Authentication is managed via context-aware WorkspaceClient initialization, including OBO (On-Behalf-Of) tokens for model serving environments to ensure queries are executed with the end-user's permissions.
- [PROMPT_INJECTION]: The skill implements an architecture designed to mitigate indirect prompt injection by using strict schema validation, guidelines for LLM judges, and explicit sanitization (escaping single quotes) for SQL queries. Evidence includes: ingestion via ResponsesAgentRequest; boundary markers through 4-6 focused evaluation guidelines; capability inventory restricted to Genie Spaces and Lakebase memory tools; and sanitization logic documented in the implementation checklist.
Audit Metadata