codex-autoresearch

Automated Codex experimentation orchestrates planning, execution, verification, and logging of iterative improvements toward measurable goals, enabling unattended overnight runs with persistent state and automatic resume capabilities.

Autonomous long-running iteration applies structured state and artifact management to execute iterative improvements across modes like loop, plan, debug, fix, and security, verifying progress through repeatable shell commands and deterministic guards.

Start Codex and describe your goal in plain language to begin the autonomous loop, using the wizard-guided interactive startup to configure modes, structured acceptance criteria, and optional background runtime.

Yes, the autonomous loop supports automatic resume and pivot handling, allowing interrupted runs to continue from their last verified state and enabling cross-run learning to improve codebases over time.

Yes, the framework supports companion repos and multi-repo scopes, managing structured state including baselines, logs, and lessons to track artifact state across all included repositories during execution.

Verification relies on repeatable shell commands and deterministic guards, meaning non-deterministic tasks or goals lacking measurable outcomes may not produce reliable progress tracking for overnight runs.