Claude-Codex Bridge

A bidirectional skill that enables AI-to-AI code comparison and review between Claude Code and OpenAI's Codex.

Features

✅ Bidirectional - Claude can call Codex, or Codex can call Claude ✅ CLI-based - No API keys or pip dependencies required ✅ Automatic truncation - Handles large contexts gracefully ✅ CLI validation - Checks for tool availability before calling ✅ Configurable timeout - Customize response times ✅ Enhanced error handling - Clear, actionable error messages

Installation

1. Copy the skill directory to ~/.claude/skills/:

   cp -r codex-compare ~/.claude/skills/

2. Ensure both CLIs are installed:

   • Codex CLI - codex command
   • Claude Code - claude command

3. Restart Claude Code to load the skill

Usage

When using Claude Code, simply ask:

• "Compare this with Codex"
• "Ask Codex to review this code"
• "What would Codex do differently?"
• "Get Codex's opinion on this"

Claude will automatically:

1. Gather the context (task, code, files)
2. Call Codex CLI
3. Present a formatted side-by-side comparison

Architecture

codex-compare/
├── SKILL.md                  # Main skill documentation
├── scripts/
│   ├── call_codex.py        # Call Codex from Claude
│   └── call_claude.py       # Call Claude from Codex
└── references/
    └── prompting-guide.md   # Best practices for AI prompting

Examples

Password Validator Comparison

# Claude's approach
def validate_password(password):
    # Simple, readable implementation
    ...

# Codex's feedback
# - Suggests configurability
# - Recommends input validation
# - Proposes using string.punctuation

Key Differences Identified

• Claude: Prioritizes clarity and simplicity
• Codex: Emphasizes configurability and robustness
• Both: Correct implementations with different trade-offs

Configuration

Input JSON format:

{
  "task_description": "Create a function to...",
  "claude_solution": "def my_function()...",
  "file_context": "Optional file context",
  "additional_context": "Optional constraints",
  "timeout": 60
}

Automatic truncation limits:

• Task description: 2000 chars
• Code solution: 5000 chars
• File context: 1000 chars
• Additional context: 2000 chars

Requirements

• Claude Code CLI installed
• Codex CLI installed
• Python 3.7+ (standard library only)
• No pip dependencies needed!

How It Works

1. CLI Detection - Validates that both CLIs are available
2. Context Gathering - Collects task, code, and relevant context
3. Truncation - Automatically truncates large inputs to prevent overflow
4. CLI Execution - Calls the target AI via subprocess
5. Error Handling - Captures failures with actionable messages
6. Formatted Output - Presents comparison in readable markdown

Development

Created using the skill-creator workflow with iterative improvements based on Codex's own review of the implementation.

Codex's Review Highlights

• ✅ Simple, clean architecture
• ✅ Works as demonstrated
• ✅ Practical improvements implemented
• ✅ Avoided over-engineering

License

MIT

Contributing

Issues and pull requests welcome! This skill was collaboratively developed by Claude and refined based on Codex's feedback.

---

Built with: Claude Code Skill Framework Reviewed by: Codex (meta!)