The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses MCP tools like execute_databricks_command and run_python_file_on_databricks to perform administrative tasks and test models on Databricks clusters. These operations are core to the skill's purpose of managing remote ML infrastructure.
[EXTERNAL_DOWNLOADS]: The documentation guides users to install widely-used Python libraries from standard registries, including mlflow, langgraph, and databricks-langchain. These downloads originate from trusted/well-known organizations aligned with the vendor.
[DYNAMIC_EXECUTION]: A code example in 4-tools-integration.md provides a custom tool using eval() for mathematical expressions. The implementation follows a common pattern of using a restricted environment by stripping __builtins__ and only allowing specific math functions.
[INDIRECT_PROMPT_INJECTION]: The skill describes an architecture where AI agents process untrusted input from the query_serving_endpoint tool.
Ingestion points: User messages enter via ResponsesAgentRequest in agent.py and query tools.
Boundary markers: The provided templates do not include explicit system-level delimiters, though they use structured request objects.
Capability inventory: The agents can potentially call tools like system.ai.python_exec and manipulate workspace resources via other MCP tools.
Sanitization: Standard model-serving guardrails are assumed; no custom sanitization logic is explicitly included in the templates.

databricks-model-serving