codex-autoresearch

Automate overnight code optimization by running a deterministic modify–verify–keep/discard loop with Codex across your repository, using autonomous goal-driven experiments to iteratively apply and evaluate changes.

A modify–verify–keep/discard loop is a deterministic process that autonomously applies code changes, verifies them against goals, and keeps successful modifications while discarding failures, enabling long-running optimization tasks across a repository.

To start an autonomous loop, describe your optimization goal in plain language and instruct the system to go, triggering the deterministic modify–verify–keep/discard workflow across your repository.

Yes, you can resume interrupted autonomous optimization runs because the system reads and writes structured artifacts like autoresearch-state.json and research-results.tsv to support session resume and cross-run learning.

Yes, the autonomous loop optimizer includes robust preflight governance with rollback options, ensuring that failed modifications from the verify phase are safely discarded from the repository.

Available modes for autonomous Codex-driven workflows include loop, plan, debug, fix, security, ship, and exec, supporting a two-phase interaction model with heavy preflight checks and rollback governance.