inno-experiment-dev

What problem does it solve?

Creates an end-to-end ML project workflow by generating an implementation plan, scaffolding project code, integrating a judge feedback loop, and submitting the final experiment run. It guides teams from idea refinement to measurable submission, ensuring reproducible experiments.

Core Features & Use Cases

• Plan generation: produces a detailed dataset, model, training, and testing plan coordinated with reference codebases.
• Code scaffolding & integration: creates a self-contained Experiment/core_code workspace with datasets, models, and training loops.
• Judge feedback loop: iterates between ML and Judge agents to refine the implementation based on atomic concepts.
• Submission handling: manages the final submission run, including checkpoint saving and result reporting.

Quick Start

Start by running the inno-experiment-dev skill after completing the code-survey and planning phases to produce an end-to-end implementation and final submission.