Validates machine learning pipelines for data leakage, training-serving skew, reproducibility issues, and model governance gaps. Use this skill whenever the user shares ML training code, preprocessing scripts, or inference code and asks for a review, audit, or "is this production-ready?" — even casual asks like "does this look right?" or "why is my model performing worse in prod?". Triggers for: "check for data leakage", "review my training pipeline", "validate my ML code", "why does my model degrade in production", or any upload of train.py / preprocessing.py / inference.py files.
