What problem does it solve?
This Skill eliminates the manual, error-prone gap between finalized experiment plans and runnable experimental results, saving ML researchers hours of repetitive coding, debugging, and deployment work for research projects.
Core Features & Use Cases
- Plan Parsing: Automatically extracts run order, milestones, GPU budget, and method details from EXPERIMENT_PLAN.md and related plan documents.
- Code Implementation: Generates reproducible training, evaluation, and baseline scripts with configurable hyperparameters and fixed random seeds.
- Pre-Deployment Review: Runs cross-model code review to catch logic bugs, incorrect evaluation metrics, and OOM risks before wasting GPU resources.
- Automated Deployment: Runs sanity checks first, then deploys experiments in parallel with automatic error recovery for common failures like out-of-memory errors or missing dependencies.
- Result Collection: Parses structured JSON/CSV output files, updates experiment trackers, and prepares initial results for downstream review workflows.
Use case: For a researcher with a finalized experiment plan for a new vertebrae segmentation method, this Skill can automatically implement all baseline, main method, and ablation experiments, deploy them to GPUs, and compile initial results without manual scripting.
Quick Start
Use the experiment-bridge skill with your refine-logs/EXPERIMENT_PLAN.md file to automatically implement, deploy, and collect results for your planned machine learning experiments.
