The Agent Skills Directory

[SAFE]: The skill implements a legitimate machine learning tuning engine. It uses PyTorch for building multi-layer perceptron (MLP) models and Optuna for Bayesian hyperparameter optimization. All core logic resides within the provided _vendor and scripts directories.
[SAFE]: Data handling is performed using standard libraries like pandas and scikit-learn. The skill reads local files (CSV/Parquet) provided via command-line arguments and performs standard data operations like splitting, normalization (StandardScaler), and missing value imputation (DNNImputer).
[SAFE]: No unauthorized network activity or data exfiltration patterns were detected. The skill interacts only with local file systems for data input and model/report output.
[SAFE]: The code is transparent and follows professional software engineering practices. No obfuscation, hidden instructions, or persistence mechanisms are present.
[SAFE]: Dependencies are restricted to well-known, reputable machine learning libraries such as torch, optuna, pandas, numpy, and scikit-learn. The use of optbinning for Weight of Evidence (WoE) encoding is a standard practice in risk modeling and credit scoring.

dnn-tuning