codex-autoresearch

Why we included this project

Anyone who already uses OpenAI's Codex CLI will find this skill useful for turning a single prompt into a long-running experiment. You give it a measurable goal, like getting a test failure count down to zero or shaving a latency number, and Codex inspects the repository, checks the plan with you, changes one thing, verifies the result, and either keeps the improvement or reverts the failure before trying again. The control script handles commits, rollback, and an append-only audit trail, so an overnight run can keep working without you watching every edit, and each trial stays a reversible Git commit. That makes it practical for teams that would rather let an automated loop chip away at test failures, coverage gaps, type errors, or warnings than fix each one by hand.