Codex Council — Development skill for Claude Code

codex-council

Independent cross-model verification and collaboration for Claude Code. Claude identifies the claims, decisions, and results worth checking, supplies the requirements and evidence, and briefs **one or more role-framed OpenAI Codex agents** to challenge them from task-specific lenses. Roles can also investigate, reproduce, test, research, or implement authorized work. Claude checks their evidence, reconciles their contributions into one result, and stays responsible for it.

The plugin is general-purpose with a programmatic center of gravity—project implementation, computer science, software and ML/AI engineering, DevSecOps, debugging/testing, and technical research—while adapting beyond those domains. Each Codex role runs on a model and reasoning effort discovered at runtime for its lens, or on your native Codex configuration. Claude itself keeps the host session's model and effort; council routing controls only the external Codex workers.

Prerequisites

• Claude Code — authenticated (claude in terminal)
• OpenAI Codex CLI — authenticated (codex in terminal)
• Python 3.10 or later on macOS or Linux (the runner uses only the standard library and POSIX process groups)

Both CLIs must be logged in and working in your terminal before using this plugin. Model discovery was verified against codex-cli 0.157.1. If a Codex build cannot answer it, discovery reports itself unavailable and roles inherit your native configuration unless you pinned a model or effort.

Install

claude plugins marketplace update
claude plugins marketplace add ehzawad/codex-council
claude plugins install codex-council@codex-council

Persists across sessions — no flags needed.

Iterating on the plugin itself? See [For development](#for-development) for the author dev loop (symlink + SessionStart hook).

Usage

/codex-council:codex-council

The plugin ships a single skill (no se