Claude–Codex Orchestrator plugin
A Claude Code plugin that hands coding tasks to Codex agents, monitors them and verifies the results independently
- 92 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- Launches Codex CLI, which writes and changes files in your repository and runs commands
- Sends prompts and project context to OpenAI and keeps event logs containing request contents
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/claude-codex-orchestratorDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Launches Codex CLI, which writes and changes files in your repository and runs commands
- Sends prompts and project context to OpenAI and keeps event logs containing request contents
This is third-party code. Review the repository files before installing.
What it does
The plugin splits roles between two models. Claude owns the plan and verifies the result, while OpenAI Codex agents, driven through its command line, take scoped implementation and review tasks. Each run keeps exact prompts, event streams and handoff records, so the work can be reconstructed. A monitor shows active agents and flags stalled ones. The plugin has four skills: `orchestrate` runs one phase, `workflow` runs the full cycle from plan to report, `report` writes `report.md` for a closed run, and `duet` asks Claude and GPT through Codex the same question and merges the answers into one. Claude settles disagreements from inspectable evidence and not by a model vote. The Python orchestrator code lives in `scripts/codex_orchestrator`.
Who it is for. Developers who have both Claude Code and Codex CLI and want one model to delegate work to the other and check it.
Good fit when
- A large task must be split: Claude plans and Codex writes the code in narrow assignments
- You need an independent review of a commit by a fresh Codex agent with no edits to the target
- Prompts, events and decisions of a run must be kept for later review
- You want one merged answer to a disputed question from Claude and GPT through `duet`
Not a fit when
- Codex CLI is not installed or its models are not available
- The task is a few lines: orchestration overhead outweighs the benefit
- You only need a second opinion with no runs and journals: `duet` alone is enough
Example request
Run `/codex-orchestrator:workflow`: Claude plans, Codex implements and a fresh Codex agent reviews the commitLimitations
- Codex CLI must be installed with access to OpenAI models, and costs of both services add up.
- Codex agents write code in your repository, so results should be reviewed and not taken blindly: the plugin itself requires independent verification of findings.
- The scripts need Python, with the version set in `pyproject.toml`.
- The monitor parses the Codex event stream format, and a format change in newer Codex versions may break it.
How to disable. In Claude Code run `/plugin uninstall codex-orchestrator@codex-orchestrator` and reload with `/reload-plugins`. Delete run folders in the project by hand.
FAQ
How do I install Claude–Codex Orchestrator?
In Claude Code: `/plugin marketplace add alexzh3/codex-orchestrator`, then `/plugin install codex-orchestrator@codex-orchestrator` and `/reload-plugins`.
How does `orchestrate` differ from `workflow`?
`orchestrate` runs one phase: execution, review, monitoring or verification within a run. `workflow` runs the whole cycle from planning to closure and report.
What does `duet` do?
Asks Claude and GPT through Codex the same question. Claude answers before reading GPT's answer, then returns one merged answer.
Why use two models?
Different model families and harnesses catch different mistakes. The authors cite work on model ensembles and ask Claude to settle disagreements from evidence and not by vote count.
Where are logs and decisions kept?
Every run keeps a journal of exact prompts, event streams and handoffs. The `report` skill writes `report.md` for a closed run.
Discussion
No questions or reviews yet.
Related
PluginsMulti-agent orchestration for Claude Code: teams, roles, autopilot and task routing across skills and subagents
Claude Code supervises while Codex, opencode and pi work on your quota: delegated edits, independent review, model discussions
A Claude Code plugin that runs one task across several external models and surfaces disagreements before you ship
A template sets up a multi-agent environment in a repository: Claude Code conducts Sonnet and Opus subagents and calls Codex CLI.
The /codex-workflows skill: Claude Code writes a script, dozens of Codex agents run it, and the run shows on a map
Workflows for Claude Code: /do, /omo, /bmad-pilot and more commands, with Codex, Gemini or OpenCode writing the code through codeagent-wrapper
README in short
The README explains the Claude and Codex division of labour and gives three reasons for two models: different mistakes, Claude's large context and native harnesses on each side. Installation, a table of the `orchestrate`, `workflow`, `report` and `duet` commands and sample prompts follow. The orchestration contract and benchmark results sit in `docs`.