What problem does it solve? When adding new datasets, models, or augmentations to a machine learning project, inconsistent architecture leads to tangled imports, manual registration boilerplate, and config drift. This Skill enforces a standardized Factory/Registry architecture so every new component integrates cleanly with Hydra configs and auto-discovery. ## Core Features & Use Cases - Registry Decorators: Register new datasets, models, and augmentations with @register_dataset, @register_model, and @register_augmentation decorators. - Config-Driven Models: Enforces a model pattern where init accepts only a Hydra cfg object and forward() returns a dict with loss, labels, and logits. - Auto-Import Discovery: Automatically imports all submodules in a directory so new files register themselves without manual import statements. - Use Case: You need to add a new Transformer model to an existing brain-decoding project. The Skill guides you to create the file in src/model_module/, apply @register_model('Transformer'), read all hyperparameters from cfg, and add a matching Hydra YAML config. ## Quick Start Ask the AI to create a new dataset class named 'custom' following the project architecture with the register_dataset decorator and a matching Hydra config file.
