A framework for evaluating AI coding agents and their skills with sandboxing, reproducibility, and data-driven analysis.
