Experiment Bridge

This bridges the gap between experiment planning and actual GPU runs. It reads your EXPERIMENT_PLAN.md, implements the code, optionally sends it to GPT-5.5 for a cross-model review to catch logic bugs before you burn GPU hours, then deploys everything and collects initial results. The sanity-first mode runs your smallest experiment first to catch setup issues early, and it'll auto-debug common failures like OOM or missing packages up to three times before bothering you. If you've got a base repo, it clones and extends it rather than starting from scratch. Useful when you've finished planning with the refine workflows and want to get from markdown to running experiments without manual translation.

Install to Claude Code

npx -y skills add wanshuiyin/auto-claude-code-research-in-sleep --skill experiment-bridge --agent claude-code