Codex Astra + Luna Orchestrator workflow
Codex Astra Orchestrator + Luna Subagents
Codex profiles where GPT-6 Astra orchestrates and reviews while GPT-5.6 Luna runs the explorer, worker, tester and researcher subagents
- 1,744 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- The installer writes configuration into the target project and appends AGENTS.md
- The profiles enable workspace-write subagents that execute code
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add workflows/codex-astra-orchestrator-luna-subagentsDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The installer writes configuration into the target project and appends AGENTS.md
- The profiles enable workspace-write subagents that execute code
This is third-party code. Review the repository files before installing.
What it does
This is a set of Codex profiles that splits roles between two models: GPT-6 Astra handles orchestration and independent review, while GPT-5.6 Luna runs the explorer, worker, tester and researcher subagents. The setup.sh or setup.ps1 installer asks your Codex plan and copies the matching profile into the target project's .codex and .agents directories, and appends orchestration instructions to AGENTS.md. On the Pro plan Astra is the root at medium reasoning; on the Plus plan Luna becomes the root to save rate limits. Variants cap concurrent subagents at two. It ships the astra-orchestrator skill, guides for different work modes and a token-usage script that reads session logs.
Who it is for. For Codex developers who want a ready multi-agent orchestration layout with roles split across models.
Good fit when
- You want Codex orchestration with a separate orchestrator, workers and reviewer
- You want to split an expensive and a cheaper model across roles to save rate limits
- You want a ready profile instead of hand-editing config.toml
Not a fit when
- You do not use Codex
- A single agent without subagents and roles is enough
Example request
$astra-orchestrator Implement the new invoice export endpoint: explorer maps the existing path first, workers make the edits, tester verifies, reviewer gives the final reviewLimitations
The profiles target Codex and its GPT-6 Astra and GPT-5.6 Luna models only; they do not fit other agents. Orchestration spends more tokens: the root holds the whole task context and each subagent carries its own. The installer edits files in the target project and, for existing files, lists the paths and asks for confirmation. Real token use depends on repo size and task, so there is no single number.
How to disable. Remove the .codex and .agents directories the installer added to the project and delete the added instruction block from AGENTS.md. For a global install, remove the corresponding settings from ~/.codex/config.toml.
FAQ
How do the Pro and Plus plans differ?
On Pro, GPT-6 Astra is the root and orchestrator; on Plus, GPT-5.6 Luna becomes the root to cut spend on the busiest role. The reviewer is Astra at low reasoning in both cases.
Does this touch my global Codex config?
By default the installer writes to .codex and .agents inside the target project. For a global install the README advises carefully merging the settings into ~/.codex by hand without overwriting existing ones.
How is Codex Astra + Luna Orchestrator different from a normal Codex setup?
These are Codex profiles that split roles between two models: GPT-6 Astra handles orchestration and independent review, while GPT-5.6 Luna does the tasks in explorer, worker, tester and researcher subagents.
How do I install the Astra and Luna orchestrator in Codex?
Clone the repository with git clone https://github.com/donvito/codex-astra-luna-orchestrator.git and connect the profiles per the instructions. Access to the needed models depends on your Codex plan.
Discussion
No questions or reviews yet.
Related
WorkflowsA Codex skill that splits work across three models: the main thread plans, Terra handles the screen, Luna runs the tests.
A Codex skill collection: phased planning, parallel subagents, documentation access, a browser, 51 hooks and ready agent roles.
oh-my-codex
oh-my-codex (OMX)
A layer on top of Codex CLI: one flow from clarifying a task to verification, agent teams, hooks, a HUD and state in .omx/
One agent launches others in the background: Claude Code, Codex, Copilot CLI, Grok Build and Pi, with a model chosen per stage.
Codex Factory runs a complex Codex project as a controlled process: plans, roles, source-bound memory and independent verification
A process kit for agents: a skill router, project memory in a DEV folder and result verification before saying done.
README in short
The README describes a Codex setup with roles split by model: GPT-6 Astra as root and reviewer, GPT-5.6 Luna as the model for execution subagents. It gives ready config.toml files for the Pro and Plus plans, variants capping subagents at two and a role table by reasoning effort. The setup.sh or setup.ps1 installer copies a profile into the project, with roles pinned to models via separate toml files. It covers the topology, tuning tips for repo size and a token_usage.py script that reads Codex session logs. Apache 2.0 license.