Use when building an active learning loop for low-label predictive tasks, where a model is trained on a small labeled set, used to score an unlabeled pool, and then used to select the next batch for annotation or review.
Use when continuously auditing model outputs to verify that they remain normal, stable, and policy-compliant over time, including sampling, drift detection, LLM-as-judge review, and storage of flagged outputs for follow-up.
Use when setting up, refactoring, or iterating on a DSPy experiment workflow, especially when you need to choose an optimizer, structure an experiment folder, name iterations consistently, or log compile/eval tokens, cost, and runtime separately. This skill is for agent-assisted DSPy experimentation rather than plain prompt-only work.