cc-codex-plugin

A Claude Code plugin that integrates OpenAI's Codex CLI for detail-oriented code review, planning, and analysis.

What is this?

This plugin gives Claude Code access to OpenAI's Codex CLI, enabling a powerful "second opinion" workflow. Codex excels at thorough, meticulous code analysis - complementing Claude's broader capabilities with sharp-eyed detail work.

Use cases:

• Deep security review of authentication code
• Bug hunting in complex algorithms
• Detailed implementation planning
• Edge case analysis
• Getting a second perspective on critical code

Prerequisites

1. Codex CLI - Install globally:

   brew install codex

2. OpenAI API Key - Authenticate:

   codex login

Installation

Install via Claude Code's plugin marketplace:

/plugin marketplace add thepushkarp/cc-codex-plugin

Or install directly from the repository:

/plugin install https://github.com/thepushkarp/cc-codex-plugin

Usage

Explicit Command

Use /codex with scoped, specific prompts:

/codex Review auth middleware for security issues. ONLY report vulnerabilities, no style suggestions. Format: file:line - issue.

/codex Find bugs in payment flow. Look for edge cases and race conditions. Skip preambles. Lead with findings.

/codex Plan database layer refactor. Output: overview, files to modify, implementation steps. No code samples.

With Options

/codex --model gpt-5.2 analyze this algorithm's complexity
/codex --sandbox workspace-write generate a test file

Prompting Tips

For best results with gpt-5.3-codex:

• Be specific - "Review for SQL injection" > "Review code"
• Set constraints - "ONLY bugs, no style issues"
• Skip preambles - "Skip preambles. Lead with findings."
• Request format - "Format: file:line - description"

Available Options

Option	Values	Default	Description
--model	any model name	gpt-5.3-codex	Model to use
--sandbox	read-only, workspace-write, danger-full-access	read-only	Sandbox mode

Automatic Delegation

Claude can also spawn Codex autonomously when it determines detailed analysis would be valuable. You'll see messages like:

"Let me have Codex take a detailed look at this code for potential bugs..."

Components

This plugin includes:

• /codex command - Direct invocation with configurable options
• codex-agent - Subagent Claude can spawn for delegated analysis
• Codex skill - Guidance on when and how to use Codex effectively

How it Works

The plugin uses Codex's non-interactive codex exec mode to run analysis:

codex exec \
  --model gpt-5.3-codex \
  --sandbox read-only \
  "<task>" \
  2>&1

The task is passed as an inline prompt argument.

Troubleshooting

"Codex CLI is not installed"

Install with npm:

brew install codex

Authentication errors

Run codex login to authenticate with your OpenAI API key.

Timeout on complex tasks

Complex analysis tasks may take time. Be patient or try a simpler task description.

License

MIT

Author

thepushkarp