Changelog

All notable changes to this project will be documented in this file.

[Unreleased]

Added codexopt improve as the one-command Codex workflow for discovery, task mining, reflective optimization, preview, and apply.
Added codexopt improve --live to opt into Codex-backed optimizer and judge runs.
Added the reflective engine for SkillOpt and GEPA inspired optimization of SKILL.md and AGENTS.md.
Added tiered rewards with verifier, judge, and static fallback modes.
Added Codex rollout parsing for codex exec --json trajectories.
Added codexopt tasks init to mine starter optimization tasks from git history, skills, and issues.
Added skillopt as a SKILL.md optimization engine with train/validation evidence splits.
Added validation-gated candidate acceptance with configurable edit budget and validation delta.
Added optional executable rollout tasks from JSON evidence.task_files.
Added temporary-repo rollout execution for candidate skills, including pass/fail artifact metadata.
Added SkillOpt metadata to optimize.json, CLI summaries, and markdown reports.
Added support for .agents/skills/**/SKILL.md discovery.
Documented progressive Codex user workflows, reflective optimization, rollout configuration, task format, artifacts, and current boundaries.

Made offline preview the default for codexopt improve so Codex and API budget are only used when explicitly requested.
Deprecated the legacy --engine gepa path in favor of the maintained reflective engine.
Updated package description to emphasize Codex and SkillOpt-style validation.

Initial open-source release of CodexOpt.
CLI workflow for init, scan, benchmark, optimize, apply, and report.
Heuristic optimization engine and optional GEPA integration path.
Run artifacts and markdown reporting.
uv-first CI pipeline for lint, test, and build.