Codex Kkp

Codex KKP

KKP: **K**otlin Claude Code s**K**ill **P**lugin.

Overview

[中文](README_CN.md) | **English**

Codex KKP is a Claude Code Plugin that enables seamless integration with Codex AI Agent for code analysis, implementation, and collaboration tasks. Provides a subagent and a skill for Codex AI Agent.

Features

• Multi-platform Support: Native binaries for macOS, Linux, and Windows
• Streamlined results: By default, only completed results are returned to reduce token consumption.
• Sandbox Modes: Configurable security levels (read-only, workspace-write, danger-full-access)
• Session Management: Resume previous sessions and maintain context
• Claude Code Subagents: Run subagents in parallel to improve performance
• Sugar-Free, Calorie-Free: Very healthy

Usage

After installation and configuration, you can use the Codex KKP plugin within Claude Code. The plugin provides a subagent and a skill for code analysis, implementation, and collaboration tasks.

You can verify successful installation by using `/agents` or by asking "What skills do you have?". Once successfully installed, you can prompt Claude Code to use Codex subagents or skills through prompts, for example:

Actively collaborate deeply with codex and subagents for task coordination and code review, ensure code quality, and perform tasks concurrently

Use Codex to review code changes

Claude Code will decide on its own whether to use Skills or agents based on the prompts.

File Operations

codex-kkp requires codex to use `sandbox=read-only` mode by default. If you need codex to directly modify files, Claude Code must specify the correct `sandbox=workspace-write` mode or add the `--full-auto` parameter (more dangerous) when calling codex-kkp.

This is mentioned in the skills progressive documentation, but if Claude Code still cannot understand it on its own, you can include these details in your prompts.

Codex will return file operation content, pro