This is an orchestrator for reproducing AI research repos in a trustworthy, auditable way. It reads the README first, picks the smallest documented inference or eval target, coordinates intake and setup, runs commands conservatively, and writes standardized outputs to repro_outputs/. The workflow is built around five helper skills that handle repo scanning, environment bootstrap, paper detail resolution, and execution reporting. It's designed for minimal smoke tests, not endless experimentation, and enforces strict patching rules so you can hand the results to another human or model for review. If you need to verify that a paper's code actually runs as documented without inventing new protocols, this is the tool.
npx skills add https://github.com/lllllllama/ai-paper-reproduction-skill --skill ai-paper-reproduction