Three-Tier Agent Orchestrator skill
A Codex skill that splits work across three models: the main thread plans, Terra handles the screen, Luna runs the tests.
- 56 ★ on the repository
- Last commit
Extended access
Why this level
- Subagents get the permissions of the Codex environment: running tests, editing files, working in browsers and apps.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/three-tier-agent-orchestratorDetects 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.
- Subagents get the permissions of the Codex environment: running tests, editing files, working in browsers and apps.
This is third-party code. Review the repository files before installing.
What it does
Three-Tier Agent Orchestrator organizes multi-agent work in Codex across three tiers. The main thread sets the goal, splits the task, accepts results and assembles the outcome. A Sol XHigh subagent takes hard analysis, deep review and debugging. Terra High or XHigh reads screenshots and drawings, works in browsers and apps, and edits text. Luna Max takes repeatable work: search, test runs, mechanical edits. The skill starts only on an explicit request: the `$three-tier-agent-orchestrator` command or its name in the prompt. Each subagent starts without the main thread history, so the goal, scope, completion criteria and verification method go into its prompt.
Who it is for. Codex users who split interface checks, tests and code analysis across models of different strength.
Good fit when
- A whole project is reviewed: Sol XHigh studies architecture, Terra High checks the interface, Luna Max runs the tests.
- Layout is fixed from a screenshot, and a separate subagent runs the regression pass right after.
- A drawing or diagram must be read visually and matched against a data table.
- Repeatable work such as batch edits and search should go to the cheapest model.
Not a fit when
- The task is short, and delegation overhead will not pay off.
- One model must own the whole result, with no context handed between subagents.
- The main model should switch automatically, which the skill cannot do.
Example request
Use $three-tier-agent-orchestrator: Sol XHigh reviews architecture and security, Terra High walks the screens, Luna Max runs tests and collects errors.Limitations
The skill targets Codex and the GPT-6.1 Sol, GPT-6.1 Terra and GPT-6 Luna models at set reasoning levels. Before delegating it checks that Sol XHigh, Terra High and Luna Max subagents launch. If any of them is unavailable, all work stops and no other model is substituted. The skill neither detects nor changes the main thread model, so the Sol role stays a recommendation. The README presents the role split as a workflow design and promises no quality per task. Write permissions, browser and desktop access come from the Codex environment. The README is in Chinese, with an English version in a separate repository.
How to disable. Delete the `three-tier-agent-orchestrator` folder from the Codex skills directory. Without an explicit call the skill does not run anyway.
FAQ
How do I install the skill in Codex?
Clone the repository into the Codex skills directory with the command below. If the skill does not appear, open a new task, then restart the Codex app.
When does the skill activate?
Only on an explicit call: `$three-tier-agent-orchestrator` or the skill name in the prompt. Questions about installing or editing the skill itself do not start delegation.
What do Terra and Luna each do?
Terra handles images, browsers, apps, text and local fixes. Luna runs repeatable, easy-to-verify operations: search, tests, mechanical changes.
Which models must be available?
`gpt-6.1-sol` at `xhigh`, `gpt-6.1-terra` at `high` and `gpt-6-luna` at `max`. Terra XHigh is checked only when the task selects it.
How are subagents named?
The task name gets a fixed suffix: `_sol_xhigh`, `_terra_high`, `_terra_xhigh` or `_luna_max`. The skill checks the suffix against the real launch parameters before starting.
Discussion
No questions or reviews yet.
Related
SkillsOne agent launches others in the background: Claude Code, Codex, Copilot CLI, Grok Build and Pi, with a model chosen per stage.
Codex Astra + Luna Orchestrator
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
A skill that hands work to a team of a CTO, five PMs and fifty engineers: roles, permissions, a task log and resource leases.
A skill that links two to four open Codex CLI windows: one splits the work while the others execute, cross-check and deliver.
A Claude Code skill that hands implementation-heavy work to Codex CLI while Claude keeps planning, review and acceptance.
Team Mode splits a Codex task between an explorer, an executor and a reviewer while the main agent keeps final acceptance
README in short
The README covers the roles, routing rules between models and the calling order. The repository holds three files: README, SKILL.md and `agents/openai.yaml`.