The Agent Skills Directory

[SAFE]: The skill provides documentation and patterns for MLflow 3 GenAI evaluation, involving standard Databricks and MLflow APIs. All functionality is consistent with the stated purpose and originates from a trusted vendor.
[SAFE]: No hardcoded credentials or sensitive data were found. Placeholders are used for configuration values like SQL Warehouse IDs, and best practices for environment-variable-based secret management are recommended.
[SAFE]: Detections of prompt injection patterns in references/patterns-datasets.md are false positives. The strings are explicitly presented as adversarial test cases within an evaluation dataset to verify agent robustness.
[SAFE]: The skill references established packages such as mlflow, openai, and pandas. Network usage for trace ingestion and OpenTelemetry exporters is standard for the documented evaluation workflows.
[SAFE]: The ingestion and processing of data for evaluation represents an indirect prompt injection surface. However, this is the primary intended use case of the skill and is handled through standard MLflow GenAI evaluation structures.
Ingestion points: Production traces and user records in references/patterns-datasets.md and references/patterns-evaluation.md.
Boundary markers: Not explicitly enforced in the provided snippets.
Capability inventory: Use of SDK/CLI for Unity Catalog and MLflow experiments.
Sanitization: Not explicitly described in patterns.

databricks-mlflow-evaluation