codex

Overview

Codex is a skill from softaworks/agent-toolkit that enables running OpenAI's Codex CLI for code analysis, refactoring, and automated editing using GPT-5.2 by default. It supports multiple sandbox modes for different security levels, session resumption for ongoing work, and configurable reasoning effort levels to balance quality and cost.

Key Features

• Multiple Model Options - Supports GPT-5.2 (flagship), GPT-5.2-max (ultra-complex), GPT-5.2-mini (cost-efficient), and other models with up to 400K input context and 128K output for software engineering tasks
• Configurable Sandbox Modes - Three sandbox levels: read-only for code analysis, workspace-write for local edits, and danger-full-access for network/broad access operations
• Session Resumption - Resume previous Codex sessions with codex exec resume --last to continue analysis or changes across multiple interactions
• Adaptive Reasoning Effort - Configure reasoning effort (xhigh, high, medium, low) to balance analysis depth with token costs for each task
• Stderr Suppression - Automatically suppresses thinking tokens (stderr) by default for clean output, with option to show them for debugging

Who is this for?

This skill is designed for developers who want to leverage OpenAI's Codex CLI through Claude Code for automated code analysis, refactoring, and editing tasks. It is particularly useful for teams that need multi-model code review capabilities, want to run automated code modifications with controlled sandbox permissions, or need to maintain long-running analysis sessions across multiple interactions.