kunkunzhishan

kunkunzhishan / codexloop

Public

A local Codex CLI loop for longer tasks with planning, review, checklist state, and audit logs.

31
1
85% credibility
Found May 20, 2026 at 31 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
JavaScript
AI Summary

CodexLoop is a session manager that helps you work with an AI coding assistant on longer, complex projects. Instead of giving the AI one giant task, you give it a goal and it breaks the work into a checklist. Before each step, the AI reviews what's been done and plans the next action. You can watch it work silently, or enable an approval mode where you see and approve each step before it runs. You can redirect the AI anytime with simple commands, and everything is saved locally so you can continue later or review what happened. It's designed for the messy middle of development—building product slices, refactoring, prototypes, and experiments where you know the direction but don't want to hand-write every next step.

How It Works

1
💡 You have a big coding project ahead

You know what you want to build but don't want to write every single step yourself. You want an AI helper that can tackle complex, multi-step tasks.

2
📋 You set up your workspace and tell CodexLoop your goal

You point CodexLoop to your project folder and describe what you want to accomplish. It creates a checklist of tasks to track the work.

3
🤖 CodexLoop starts working through your checklist

The AI examines your project, plans the next useful step, and begins making changes. You see a card showing what it's about to do.

4
You can watch quietly or stay involved
👀
Watch mode - CodexLoop works autonomously

The AI runs through tasks on its own. You can check progress anytime with a simple command.

✅
Approval mode - You approve each step

Before the AI makes any changes, it shows you its plan. You can approve it, reject it with feedback, or steer it in a new direction.

5
🔄 You can steer the work anytime

If the AI goes in an unexpected direction, you type a quick command to redirect it. The checklist updates to match your new priorities.

6
📊 You check progress and see what changed

After each loop, you see a summary of completed tasks, new files created, and what's left to do. Everything is logged for later review.

🎉 Your project is complete or ready for review

CodexLoop finishes when the checklist is done or you hit your loop limit. Your work is saved, audited, and ready for you to use.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 31 to 31 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is codexloop?

CodexLoop wraps the Codex CLI in a persistent control loop for longer tasks. Instead of dumping a massive prompt and hoping for the best, it maintains a living checklist, reviews progress between steps, and lets you steer or approve each action before it runs. Everything lives locally under `.agent2long` in your workspace, giving you an audit trail you can inspect at any time. Built in JavaScript, it runs as a console session where plain text is ignored and you navigate entirely with slash commands like `/status`, `/checklist`, `/steer`, and `/approve`.

Why is it gaining traction?

The killer feature is human-in-the-loop approval. You can set `--approval` mode and see exactly what Codex plans to do before it touches your files, then approve, reject with feedback, or steer in a new direction. This addresses the "drift" problem that plagues long Codex runs. The checklist gives you visibility into what the agent thinks remains, and the audit logs under `.agent2long` mean you can always reconstruct what happened and why. For teams running autonomous Codex workflows, this adds a safety layer without killing automation entirely.

Who should use this?

Backend or fullstack developers tackling multi-step features, refactors, or prototypes where a single Codex call would be too large to monitor. Product teams prototyping vertical slices who want checkpoints before the agent diverges. Anyone running Codex locally who needs an auditable, steerable workspace rather than a black box. If your task fits in one prompt, plain Codex CLI is faster; if it spans hours or dozens of steps, CodexLoop keeps you in the loop.

Verdict

CodexLoop fills a real gap for local Codex users who need oversight without cloud dependencies. With 31 stars and early-stage maturity, it is functional but unproven at scale. The 0.85% credibility score reflects a small community and limited production battle-testing. Worth trying for serious local Codex workflows, but treat it as experimental and review the audit logs carefully.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.