/sol: Claude plans, Sol implements, Claude reviews
A Claude Code skill for multi-model AI coding: Claude writes the task brief and reviews the finished diff, while GPT-6 Sol implements the change through the OpenAI Codex CLI, so the same model never grades its own work.
This skill splits coding work across two different models from two different labs so that the model that wrote a change is never the one that approves it. Claude, running as the user's own Claude Code session, writes a brief for the task; GPT-6 Sol, invoked at high reasoning effort through the OpenAI Codex CLI, implements the change and adds tests; Claude then reviews the actual diff, reruns the checks itself, and directs any corrections.
Installation adds the Codex CLI, then the skill through a Claude Code plugin marketplace; roughly fifty other agent hosts, including Cursor, Copilot and Gemini CLI, can install it through a separate skills installer. The authors note that this workflow is slow, citing an eight-minute run for a one-file change.