#341341st of 455 in Skills, plugins and prompts
codex-autoresearch
Codex skill that loops on a metric, keeping or reverting each change
- Stars
- 3.2k
- License
- MIT
- Last commit
- Sep 2026
- Last release
- Jun 2026
- Language
- Python
1,388 on 22 Apr 20263,196 on 11 Nov 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
A skill for OpenAI Codex that runs an experiment loop in a Git repo. Each iteration changes one thing, commits, and runs a metric command. The commit is kept only if the number improves and an optional guard passes, otherwise it is reverted with git revert. It runs in the foreground via Codex Goal or as a detached controller, and logs history to autoresearch-results/events.jsonl.
Who it is for: Codex users who want automated, metric-driven code improvement loops
Strengths
- Every trial is a Git commit; non-improving trials or failed guards are reverted
- Append-only events.jsonl history exports to TSV or a static HTML report
- Stops with exact error and log path on out-of-scope edits, bad metrics or timeouts
- Accepts any metric a command prints as one number or a named JSON key
Weaknesses
- Needs Codex running with Full Access (--dangerously-bypass-approvals-and-sandbox) by default
- Requires a clean named Git branch; one run manages one repository only
- Works only with Codex, not other coding agents
- Foreground mode needs a current Codex release with the Goal feature
What it needs
- no GPU
- Needs Codex CLI, Git
- Models: OpenAI Codex
What changed
- v0.5.0 ↗ (opens in a new tab)Breaking
“For the best long-running experience, start Codex with:”
From the release notes. No flag means the note does not say so, not that the upgrade is safe.
Also in Skills, plugins and prompts
See all 455| Rank | Project | Score |
|---|---|---|
| 1 | claude-memPersistent cross-session memory plugin for Claude Code and other coding agents | 97 out of 100 |
| 2 | impeccableDesign skill with 24 slash commands for AI coding agents' frontend output | 97 out of 100 |
| 3 | gstackMarkdown slash-command skills that give Claude Code planning, review, QA and ship roles | 96 out of 100 |
| #4 | rufloAgent harness adding swarms, memory and hooks to Claude Code and Codex | 96 out of 100 |
| #5 | cavemanShortens coding-agent replies and compresses tool output before the model reads it | 95 out of 100 |