wh1tespirit

Codex Orchestration — Development skill for Claude Code

Development community

Agent skill: a lead agent plans and verifies, Codex workers research and write the code, then it reviews, commits, merges into dev/main and cleans up.

How to install Codex Orchestration

This entry records only its repository, not the path inside it, so there is no exact command to give. Open wh1tespirit/codex-orchestration and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Codex Orchestration does

Agent skill: a lead agent plans and verifies, Codex workers research and write the code, then it reviews, commits, merges into dev/main and cleans up. Works with Claude Code, Codex, OpenCode, Hermes and other SKILL.md agents.

Alternatives in Development

Ralph Loop — Autonomous coding loop — works through tasks, commits, repeats 14k ★
Evolving Agents — A director agent that learns how your team works: it plans by walking your CLAUDE.md and memory as linked stat 453 ★
Install Global — Install global Claude config — merges into existing ~/.claude/ without overwriting 335 ★

README

codex-orchestration

An agent skill that turns your main AI agent into a **lead** and sends the code work to cheaper **Codex workers**. You give it an approved plan. It researches, implements, verifies, reviews, commits, merges into `dev` (or `main`), and removes the temporary branch and worktree. You get a clean repository and a short report.

The skill follows the open [Agent Skills](https://agentskills.io/) format (`SKILL.md`). It is written for Claude Code, Codex CLI, OpenCode, Hermes Agent, and any other agent that loads `SKILL.md` skills and can run shell commands.

**Status:** version 0.1.0. The skill text is complete, but it has not yet been run end to end on a real project, and it has not been tested in every agent listed above. Treat the first run as a trial and read the report.

What it does

approved plan
  → baseline (git status, dev branch?)
  → Researcher: reads and searches the code (read-only)
  → Implementer: writes the code, delegates small lookups to its own subagents
  → lead verifies: touched files, diff, project checks
  → fresh reviewer checks the change (up to 2 fix rounds)
  → lead accepts against the plan's definition of done
  → lead commits the task files
  → merge --no-ff into dev (or main), remove worktree and branch
  → report

Role	Default model	Default effort	Job
Lead	your current agent	—	Plans assignments, verifies, commits, merges. Does not read or write code.
Researcher	`gpt-6-luna`	`xhigh`	Finds files and code paths. Read-only.
Implementer	`gpt-6.1-sol`	`high`	Writes the code. Hands lookups to subagents.
Reviewer	set by your review skill	—	Checks the change without seeing the plan.

Why use it

Lower cost. The most expensive model decides and checks. It does not spend tokens on searching and typing code.
Independent checks. The lead never trusts a worker's own report. It re-runs the project checks and reads the diff. A reviewer with