ming-bench

Compare two benchmark runs side-by-side with narrative analysis of struggles, cost, and gate friction.

Record a 5W1H changelog entry for the latest framework change into DESIGN_EVOLUTION.md. Run after committing changes to capture who/what/when/where/why/how.

Analyze how well an agent followed its CLAUDE.md strategy rules during a benchmark run. Use when analyzing agent behavior, rule compliance, or debugging why an agent failed.