experiment-bridge

To automate running experiments from a markdown plan on GPU, you provide an EXPERIMENT_PLAN.md file. The skill parses this file to automatically implement, deploy, and run the planned experiments while collecting initial metrics and logs.

To ensure reproducible results when deploying automated experiments, the workflow fixes random seeds during execution. It also exports all collected metrics and outputs in structured JSON and CSV formats for later consistent review.

The best way to convert an experiment plan into initial metrics automatically is by parsing EXPERIMENT_PLAN.md. The process bridges idea discovery to execution by deploying the plan on GPU and collecting initial results with exported logs.

No, you do not need to write deployment code manually to run GPU experiments from a planning document. You simply provide the plan file, and the automation workflow handles the implementation, deployment, monitoring, and metric collection.

Experiment outputs for later review and analysis should be in JSON and CSV formats. The automated workflow saves collected initial metrics and logs in these structured formats to enable reproducible results and quick evaluation.

Yes, you can use this workflow to reduce manual debugging time when testing new ideas. It bridges the gap between discovery and execution by turning brainstormed plans into runnable experiments, minimizing manual handoffs and debugging.