The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill utilizes dynamic Python scripts and subprocess calls to facilitate agent evaluation. Specifically, scripts/create_dataset_template.py and scripts/run_evaluation_template.py generate boilerplate code to run the user's agent. scripts/validate_tracing_runtime.py uses importlib.import_module to load and test the agent entry point. These are standard integration patterns for evaluation tools.
[INDIRECT_PROMPT_INJECTION]: The skill inherently processes external data (evaluation test cases) through an LLM agent. While this presents an indirect prompt injection surface, the skill implements best practices by providing structured dataset templates and encouraging the use of dry runs and validation scripts to detect anomalies before full-scale execution.
[EXTERNAL_DOWNLOADS]: The skill refers to official MLflow documentation and instructs the user to install standard, well-known Python packages (e.g., mlflow, databricks-sdk, ragas) from public registries. All network references target trusted domains like mlflow.org.
[COMMAND_EXECUTION]: Several scripts use subprocess.run to interact with environment tools like the Databricks CLI and the MLflow CLI (mlflow doctor). These calls are used for diagnostic and configuration purposes and do not take unvalidated user input.

agent-evaluation