‹ 首页

codex-agent-collaboration

@fortescarlet · 收录于 1 周前

Delegate coding tasks to Codex AI for implementation, analysis, and alternative solutions. Use when you need a second AI perspective, want to explore different approaches, or need specialized Codex capabilities for complex coding tasks.

适合你,如果需要在编码时借助 AI 获得不同实现思路或替代方案。

/ 下载安装
codex-agent-collaboration.skill双击,或拖进 Claude 桌面版 / Cowork,即完成安装↓ .skill↓ .zip
用别的 agent?下载 .zip 解压,把文件夹放进它的技能目录
Claude Code~/.claude/skills/(项目级 .claude/skills/)
Codex CLI~/.codex/skills/
Cursor自动读取上面两处目录
其他工具见其文档的「skills」目录;两个下载是同一份文件,只是名字不同
/ 通过 npx 安装 校验哈希
npx oh-my-skill add fortescarlet/codex-kkp/codex-agent-collaboration
/ 通过 bash 安装
curl -fsSL https://oh-my-skill.com/install.sh | bash -s -- fortescarlet/codex-kkp/codex-agent-collaboration
/ 已经装过?验证本机副本,不用重装
npx oh-my-skill verify fortescarlet/codex-kkp/codex-agent-collaboration
安装目标可用 --agent / --scope 或 --to 明确指定;省略时只会在唯一已存在的 agent 目录上自动选择,零命中或多命中会停止并提示。content_hash 缺失或不一致均拒装。
25GitHub stars
~720最小装载
~2.1K含声明引用
~2.1K文本包总量
镜像托管

怎么用

技能原文 SKILL.md作者撰写 · MIT · c829e64

Codex CLI Skill

This skill enables Claude Code to execute tasks using OpenAI's Codex AI agent.

Overview

The codex-kkp-cli is a Codex Agent CLI tool, allowing you to:

  • Execute coding tasks and get implementations
  • Perform code analysis and reviews
  • Get alternative solutions and suggestions
  • Collaborate with Codex for cross-checking implementations
Usage
Basic Syntax
# Direct call with platform-specific executable
executables/codex-kkp-cli-{platform} --cd=/absolute/path/to/project [options] "<task_description>"

Where {platform} is one of:

  • macosx64 - macOS Intel (x86_64)
  • macosarm64 - macOS Apple Silicon (ARM64)
  • linuxx64 - Linux x86_64
  • linuxarm64 - Linux ARM64
  • mingwx64 - Windows x86_64

Platform Auto-Detection Helper: A platform detection script is provided to help identify your current platform:

On Windows, Just use mingwx64 platform directly, no need to use script detection.
# Unix/Linux/macOS
codex-kkp-cli-platform
# Outputs: macosx64, macosarm64, linuxx64, or linuxarm64
communication

This is AI-to-AI communication between You and Codex. PRIORITIZE ACCURACY AND PRECISION over human readability. Use structured data, exact technical terms, full paths, and precise details. NO conversational formatting needed.

Required Parameters

| Parameter | Description | |--------------|------------------------------------------------------------| | Task | The task description (positional argument, must be quoted) | | --cd=<dir> | Working directory (ABSOLUTE PATH REQUIRED) |

Optional Parameters

| Parameter | Description | |--------------------------------|--------------------------------------------------------------------------------| | --session=<id> | Session ID (STRONGLY RECOMMENDED for follow-up chats to maintain context) | | --sandbox=<mode> | Sandbox mode. Default is read-only. See [sandbox-modes.md](sandbox-modes.md) | | --full-auto | Allow Codex to edit files automatically | | --image=<path> | Include an image file (ABSOLUTE PATH, can repeat) | | --skip-git-repo-check[=BOOL] | Skip Git repository check. Default is true. Use =false to enable Git check |

For output options (--full, --output-last-message, --output-schema), see [outputs.md](outputs.md).

NOTE that parameters and values are connected by an EQUAL SIGN =, not a space.

Response Format

Returns JSON with "type": "SUCCESS" or "type": "ERROR".

{
  "type": "SUCCESS",
  "session": "xxxxxxx",
  "content": {
    "agentMessages": "I've analyzed the code and found...",
    "fileChanges": [...],   // Optional
    "nonFatalErrors": [...] // Optional
  }
}
  • fileChanges and nonFatalErrors is nullable.
  • Error responses do NOT include a session field.
Quick Example

New Session:

executables/codex-kkp-cli-{platform} --cd=/path/to/project "Explain the main function in Main.kt"

Continue Previous Session:

executables/codex-kkp-cli-{platform} --cd=/path/to/project --session=xxxxxxx "Explain the main function in Main.kt"

More examples: [examples.md](examples.md)

按 MIT 许可原样转载,未经改动 · 在 GitHub 查看 →

评论

登录即可评论;带「已验证安装」的,是发布者名下有本店的安装或持有记录。