Which AI models does the Codex skill use?

By default, it uses GPT-5.2, but it also supports GPT-5.2-max for ultra-complex reasoning, GPT-5.2-mini for cost-efficient coding, and GPT-5.1-thinking for deep analysis.

Can I continue a previous Codex session?

Absolutely. By using the 'codex resume' command, the skill pipes your new prompt into the previous session, maintaining the same model and context for iterative work.

What is the Codex skill for Claude Code?

The Codex skill is a specialized extension that enables Claude Code to interface with the Codex CLI, allowing it to use GPT-5.2 models for advanced software engineering tasks like automated editing and complex problem-solving.

How do reasoning effort levels work?

Reasoning effort levels (low, medium, high, xhigh) allow you to control how much cognitive resources the AI spends on a task. Use 'low' for simple formatting and 'xhigh' for complex architectural changes.

Is the Codex skill safe for my local files?

Yes, the skill uses sandbox modes. You can set it to 'read-only' for analysis or 'workspace-write' when you want the AI to apply edits to your local codebase.

Codex AI Engineering

Name: Codex AI Engineering
Author: softaworks

bysoftaworks

•

347

Herramientas para Desarrolladores

Integrates the Codex CLI and GPT-5.2 models to automate complex software engineering tasks including refactoring, bug fixing, and deep code analysis.

Acerca de

The Codex skill empowers Claude Code with state-of-the-art software engineering capabilities by leveraging the Codex CLI and the GPT-5.2 model family. It allows developers to execute sophisticated coding tasks with adjustable reasoning efforts—ranging from quick fixes to ultra-complex architectural changes—while maintaining strict control through configurable sandbox environments. With native support for SWE-bench optimized models and seamless session resumption, this skill is designed for high-impact development workflows that require more than basic code generation.

Características Principales

Native support for GPT-5.2 models with 76.3% SWE-bench performance
Adjustable reasoning effort levels (xhigh, high, medium, low) for task-specific optimization
Granular sandbox controls including read-only and workspace-write modes
Seamless session resumption to continue complex coding tasks without losing context
Automatic stderr suppression for cleaner output and focused agent interaction
347 GitHub stars

Casos de Uso

Automating large-scale codebase refactoring and architectural migrations
Performing deep-reasoning security audits and performance optimizations
Executing autonomous bug fixes with full-auto workspace editing

Acerca de

Características Principales

Native support for GPT-5.2 models with 76.3% SWE-bench performance
Adjustable reasoning effort levels (xhigh, high, medium, low) for task-specific optimization
Granular sandbox controls including read-only and workspace-write modes
Seamless session resumption to continue complex coding tasks without losing context
Automatic stderr suppression for cleaner output and focused agent interaction
347 GitHub stars

Casos de Uso

Automating large-scale codebase refactoring and architectural migrations
Performing deep-reasoning security audits and performance optimizations
Executing autonomous bug fixes with full-auto workspace editing