experiment-bridge

Yes, you can run staged machine learning workflows by executing a small sanity experiment first, then deploying the full milestone sequence. This approach includes checkpointing and result collection to ensure safe, structured experiment progression.

Automated experiment planning requires reading EXPERIMENT_PLAN.md plus supporting proposal and tracker files, then implementing the specified training and evaluation logic. The Skill extracts workflow orchestration details to generate reproducible scripts and deploy them safely.

Pre-deploy code review for machine learning experiments catches correctness issues and metric-ground-truth mistakes before GPU execution. The Skill optionally runs a cross-model review to validate evaluation metrics and code accuracy prior to deployment.

To convert an experiment plan into executable training code, the Skill reads EXPERIMENT_PLAN.md and supporting tracker files to implement reproducible scripts exactly as specified. It bridges written plans into working training and evaluation pipelines ready for GPU deployment.

Deploying machine learning training scripts to a GPU involves implementing the code from your plan, verifying ground-truth-based evaluation, and safely executing the pipeline. The Skill manages this deployment while saving results in parseable formats for downstream review.

Yes, you need an EXPERIMENT_PLAN.md file along with supporting plan, tracker, and proposal files to generate training and evaluation code. The Skill reads these documents to implement reproducible scripts and deploy them for initial result collection.