Worked on the probabl-ai/skore repository to enhance the robustness of ROC-AUC metric reporting for machine learning estimators. Addressed a recurring issue where estimators lacking the predict_proba method caused runtime errors by introducing a defensive guard, _check_roc_auc, within the API. This solution ensured that ROC-AUC metrics were only exposed when supported, preventing AttributeError and improving reliability for users evaluating diverse models. Applied defensive programming techniques in Python, focusing on API design and maintainability. The update reduced support overhead and defects related to metric reporting, enabling more consistent and stable evaluation workflows across a broader range of estimator types.