Wraps the Codex CLI to give Claude direct access to GPT-5.2 for code analysis, refactoring, and automated edits. It handles the model selection dance (asking about reasoning effort levels), picks the right sandbox mode for read-only reviews versus live edits, and suppresses thinking tokens by default so you get clean output. The resume functionality is handy since you can pick up previous sessions with a simple prompt. Honestly feels a bit meta having Claude orchestrate a competing model, but if you're already using Codex CLI and want it integrated into your Claude workflow, this handles all the flag wrangling and safety checks so you don't have to context switch to your terminal.
npx skills add https://github.com/softaworks/agent-toolkit --skill codex