Using Data Analysis Skills

Pick the narrowest skill that covers the current task. When the work spans multiple lifecycle stages, start from the orchestrator and let it call the others.

Situation	Skill
An end-to-end project or an ambiguous modeling request	running-decision-grade-data-science
Data meaning, joins, labels, or ground truth may be untrustworthy	auditing-data-and-ground-truth
Designing splits, feature eligibility, baselines, or comparisons	designing-leakage-safe-experiments
A metric dropped, results disagree, or training-serving mismatch	diagnosing-ml-failures
Reviewing whether results support a claim or launch decision	validating-models-and-claims
Packaging finished work for independent reproduction	shipping-reproducible-results

Each skill states its own non-goals in its description; respect them. The orchestrator running-decision-grade-data-science already routes to the other five at the right lifecycle stage, so do not stack it with them manually for the same step.