# Codex Delegate

A Claude Code skill that delegates implementation-heavy coding work, such as multi-file refactors and boilerplate generation, to the Codex CLI as an execution specialist, while keeping planning, review and final acceptance inside Claude.

Codex Delegate follows a fixed pattern: Claude writes a task file describing scope and constraints, launches Codex synchronously through a wrapper script, the wrapper emits sentinel files and a machine-readable result file once Codex finishes, and Claude then reviews the resulting diff and runs its own verification before accepting the change; wrapper success alone is never treated as final acceptance. It is intended for tasks that consume many tokens but require little judgment, such as multi-file implementation, mechanical refactors, boilerplate generation, test scaffolding and large batch edits, and explicitly not for architecture decisions, root-cause debugging, security review or ambiguous product decisions.

The skill is positioned as a lighter-weight alternative to OpenAI's own official Codex integration for Claude Code: a single synchronous skill with no job tracking rather than a persistent broker process with background jobs and slash commands, using bash and PowerShell wrappers with no Node.js runtime required.

It is written in Python and released under the MIT license.

Website: https://github.com/WenyuChiou/codex-delegate