The Agent Skills Directory

[SAFE]: The skill uses official, versioned libraries from Databricks and MLflow (databricks-openai, databricks-sdk, mlflow) to manage agent orchestration and distributed tracing.
[SAFE]: Implements a multi-turn approval flow for MCP server tools (Unity Catalog connections or apps), ensuring that sensitive tool calls require explicit user confirmation.
[SAFE]: No hardcoded credentials or sensitive file paths were detected; the skill relies on environment variables and workspace-level permissions managed through service principals.
[PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection as it ingests untrusted user input and passes it to a server-side agent loop capable of executing various tools.
Ingestion points: request.input in the invoke_handler and stream_handler functions within agent_server/agent.py.
Boundary markers: None explicitly defined in the provided code snippets to distinguish between instructions and user-provided data.
Capability inventory: Ability to execute Unity Catalog functions, query Genie spaces, access Knowledge Assistants, and call MCP server endpoints.
Sanitization: No input validation or sanitization logic is present in the skill instructions to filter potentially malicious instructions within user messages.

supervisor-api