Give Codex a programming task and have it automatically delegate the file editing to Claude Code, then review the result before responding to you.
Speed up coding workflows by combining Codex's planning and review strengths with Claude Code's file-editing capabilities in a single automated loop.
Install the skill with one npx command and start using it immediately inside an existing Codex session without manual configuration.
Use the collaboration loop for tasks like refactoring or feature implementation where planning and editing are naturally separate steps.
Requires both OpenAI Codex and Claude Code installed locally on the same machine before installing the skill.
This repository is a skill, meaning a plugin, that makes two AI coding assistants work together in a coordinated way: Codex from OpenAI and Claude Code from Anthropic. Each assistant takes a different role. Codex does the planning, hands the actual coding work to Claude Code, and then reviews what Claude produced before reporting back to you. The goal is to combine Codex's planning and review abilities with Claude Code's file-editing capabilities in a single automated loop. When you invoke the skill inside a Codex session and give it a programming task, Codex decides whether the task is a good candidate for delegation. If it is, Codex launches Claude Code in the background, captures the output while Claude edits files in your workspace, and then reads through the changes before responding. This keeps Codex responsible for the final result while Claude handles the hands-on implementation. Installation uses a single terminal command with npx, the standard Node.js package runner, which places the skill files in the correct location inside your Codex installation. You can also clone the repository manually if you prefer. After installing, you restart Codex and invoke the skill by name in your next message. The skill is built for local use where both Codex and Claude Code are installed on the same machine. It skips per-tool approval prompts for Claude's edits within the shared workspace, which makes the handoff faster, but the README is explicit that this does not bypass credentials, production system access, or destructive commands. The safety boundary applies only to local workspace file edits.
← ashinno on gitmyhub — every repo by this author, as a profile.
Verify against the repo before relying on details.