skill-test

Automated Databricks skill evaluation applies deterministic and MLflow-based metrics to measure WITH-skill versus WITHOUT-skill performance differences. The framework executes per-skill evaluation runs, reads frontmatter metadata, and generates quality gate results across multiple environments.

The GRP workflow performs ground-truth generation, candidate review, and promotion to optimize Databricks skills. It generates versioned baselines, executes regression checks, and produces optimized SKILL.md variants to promote validated improvements.

Yes, multi-environment evaluation supports per-skill manifests and configurable criteria for regression tracking. The framework reads frontmatter metadata, executes per-skill evaluation runs, and generates versioned baselines to scale regression checks across multiple skills.

Ground-truth generation creates JSON assertions and regex pattern checks for deterministic skill evaluation. The framework applies these assertions during per-skill evaluation runs, then routes candidates through a review and promotion workflow.

Yes, the Databricks SDK, MLflow, and PyYAML dependencies are required to execute the evaluation framework. These dependencies enable automated testing, trace analysis, and regression tracking across SKILL.md artifacts within Databricks environments.

Measuring WITH-skill versus WITHOUT-skill results provides direct effectiveness comparison using MLflow evaluation and deterministic assertions. This framework supports focused trace analysis, configurable criteria, and versioned baselines to validate improvements quantitatively.