Codex Astra Luna Orchestrator 🤖 Agent Open source

What it does

This project is a ready-made configuration for OpenAI Codex that splits work between a stronger model and a cheaper one. A root orchestrator plans the task and hands bounded pieces to named subagents. A separate reviewer role checks the result at the end. The installer asks which Codex plan you are on and copies the matching profile into your project. On Pro, GPT-6 Astra orchestrates and GPT-5.6 Luna does the execution work. On Plus, Luna does both jobs to save usage. The Astra reviewer is kept on both plans.

What is inside

• Four profiles: Pro and Plus, each with a variant that limits you to two subagents at a time
• Role files: explorer, worker, tester and researcher pinned to Luna, and a reviewer pinned to Astra at low reasoning
• An astra-orchestrator skill that you call with $astra-orchestrator to run the explore, implement, test and review loop
• AGENTS.md instructions that are appended to your project's file without overwriting what is already there
• Guides for fast iteration, complex repositories, routine coding and token use
• token_usage.py, which reads Codex session logs and reports usage by thread, role and model

Works with

OpenAI Codex CLI, the IDE extension and the desktop app, since they all share the same .codex configuration.

How to install

Clone the repository, then run the installer and point it at your target project:

git clone https://github.com/donvito/codex-astra-luna-orchestrator.git
cd codex-astra-luna-orchestrator
./setup.sh

Maintenance and safety

The project is Apache-2.0 licensed and was updated recently. Setup asks before each part is installed. It lists any file it would overwrite, and the default answer is no. It never pushes or changes your global config. The model names and plan limits are tied to OpenAI's current lineup, so profiles will need edits when models change. Running several subagents uses noticeably more of your quota than a single session.

Who should use it

It suits Codex users who want multi-agent orchestration without writing role files themselves, and who want to see what that costs in tokens. If you use a different coding agent, this is of little use.