Can I run Codex and Claude tasks at the same time?

Yes, the skill is designed for parallel work, allowing you to design the next system component while Codex drafts the current implementation.

How does the token economics benefit work?

Claude's output tokens are more expensive than input tokens. By having Codex write directly to files, you only pay for the input tokens required to read the summary or diff, saving up to 80% on generation costs.

Why use Codex as a coworker instead of Claude alone?

Codex tokens are significantly cheaper than Claude's output tokens. Delegating implementation to Codex allows you to save costs while utilizing Claude for high-level reasoning and integration.

What is the 'reasoning effort' setting?

It allows you to match the model's intensity to the task: 'medium' for boilerplate, 'high' for standard implementation, and 'xhigh' for complex debugging or architecture.

Is the code generated by the coworker safe to use?

The skill follows a 'Trust but Verify' model, encouraging the use of automated quality gates like type-checking, linting, and git diff checks to validate all output.

Codex Coworker

Name: Codex Coworker
Author: phrazzld

byphrazzld

•

生産性とワークフロー

Delegates coding tasks to the Codex CLI for cost-effective implementation, parallel brainstorming, and automated code reviews.

The Codex Coworker skill transforms Claude into a project lead by offloading routine implementation, unit testing, and boilerplate generation to the Codex CLI. By leveraging separate, cheaper token pools and parallel execution, this skill allows users to maintain architectural focus while the 'coworker' handles the heavy lifting of code production. It includes pre-built templates for delegation, reasoning effort controls for different task complexities, and integrated quality gates to ensure all generated code meets project standards through automated linting and type-checking.

主な機能

01File-based output workflows to minimize input token consumption

02Token optimization reducing code generation costs by approximately 80%

03Granular reasoning effort controls (medium to xhigh) for task-specific logic

04Parallel task execution for simultaneous design and implementation work

052 GitHub stars

06Pre-delegation checklists and quality gates for automated validation

ユースケース

01Implementing CRUD operations and boilerplate from a defined architectural pattern

02Conducting secondary code reviews and brainstorming alternative logic approaches

03Generating comprehensive test suites for existing functions while you plan the next feature

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add phrazzld/claude-config codex-coworker

For use in Claude.ai and ChatGPT

Download Skill