databricks-unity-gateway
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill provides comprehensive and secure guidelines for interacting with Databricks services. It explicitly recommends against insecure practices like passing secrets in plaintext via CLI arguments, instead providing shell script templates that use environment variables and secured temporary files (
chmod 600) to protect sensitive API keys.- [INDIRECT_PROMPT_INJECTION]: The skill provides an interface for the agent to construct and execute configuration payloads based on user instructions. - Ingestion points: User-provided parameters for model services, MCP services, and budgets ingested into JSON payloads and CLI commands (found in references/budgets.md, references/mcp-services.md, etc.).
- Boundary markers: The instructions emphasize using structured JSON for payloads, which acts as an inherent schema boundary.
- Capability inventory: The skill enables execution of Databricks CLI commands for resource management and Python SDK calls for model inference.
- Sanitization: The documentation provides specific mitigations for handling secrets, including the use of environment variables and the
mktemputility to manage sensitive input safely (references/model-provider-services.md).
Audit Metadata