What problem does it solve? Research experiments often fail to produce trustworthy results because code architecture, configs, seeds, and outputs are not recorded consistently. This Skill turns a paper claim or experiment idea into a structured code architecture, run conventions, and thesis-console records so every run can be traced, reproduced, and mapped to a claim. ## Core Features & Use Cases - Experiment Architecture Planning: Separates data pipelines, model/algorithm modules, training, evaluation, metrics, and reporting with explicit contracts and boundary rules. - Run and Reproducibility Conventions: Defines stable experiment IDs, config/output directory layouts, machine-readable metrics, environment snapshots, and reproducibility status labels. - Multi-Target Execution Strategy: Coordinates local Mac smoke tests, a remote RTX 4060 desktop as the primary GPU target, and AutoDL cloud as a fallback with auto-save evidence and automatic shutdown. - Use Case: A graduate student needs to implement a baseline and ablation for a thesis claim. The Skill produces the module layout, config files, train/evaluate commands, an experiment registry row, and a remote GPU runbook before any training starts. ## Quick Start Use the research-experiment-engineering skill to plan the code architecture and runbook for my next thesis experiment.
