mirror of
https://github.com/yamadashy/repomix.git
synced 2026-09-19 06:00:49 +08:00
fc3b5a39ce
decision(perf-testing): take the fastest of five runs rather than a single sample — noise can only slow a run down, so the minimum is the stable estimator and removes the CI flake risk a one-shot measurement carried Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>