What problem does it solve? Research code often contains hard-coded paths, missing seeds, and undocumented configs that make experiments irreproducible and thesis claims unverifiable. This Skill enforces experiment contracts, smoke tests, and machine-readable output manifests before expensive GPU runs. ## Core Features & Use Cases - Experiment Contract Validation: Checks that every formal run has a config, seed, split, metric, output path, registry row, and smoke config via check_experiment_contract.py. - Project Skeleton Generation: Renders a standard src/configs/tests layout with render_project_skeleton.py for new research projects. - Remote GPU Run Templates: Provides sync/run/fetch/archive templates for a remote 4060 desktop and AutoDL fallback with environment snapshots, checksums, and auto-shutdown. - Use Case: Before launching an overnight training run on a remote 4060 GPU, use this Skill to verify the experiment contract, generate an environment snapshot, and confirm the output manifest will contain metrics, logs, and resolved config. ## Quick Start Use the research-code-quality skill to check my experiment code and config contract before I start GPU training.
