I made a tiny Codex Desktop skill for switching to Luna Max

Codex Desktop currently doesn't offer a Luna Max setting in its model selector, although Codex CLI and the Codex iOS app do. I wanted to use it in Desktop too. Asking Codex Desktop to “switch this chat to GPT-5.6 Luna with max reasoning” can get you a flat “I can't change my own model.” You need a longer instruction explaining which thread tools to call and what model and reasoning settings to pass. So I put those instructions into a tiny skill called luna-max . Once installed, I invoke $luna-max in the chat. When I invoke $luna-max , it: finds the active Codex thread for the current workspace; sends one internal request for gpt-5.6-luna with max reasoning; and leaves repository files, provider configuration, and unrelated threads alone. The skill spells out the steps, including asking you to choose if it can't identify the current thread unambiguously. It's a small convenience, but saves typing the detailed instructions each time. Mine targets Luna 5.6 with max reasoning , because I personally prefer 5.6. You can easily adapt it for Luna 6.1 by changing the model ID and matching wording throughout the file to the version available in your app. It's just one small skill file, using Codex Desktop's thread tools. Save it as ~/.codex/skills/luna-max/SKILL.md , then invoke $luna-max . This is an unofficial skill and depends on those tools being available in your Desktop session. Source and installation — my collection of useful agent skills . Contributions and fixes welcome. One UI quirk in Codex Desktop: the model selector still shows Luna Medium after the switch, even though the session records Luna Max . You can inspect the recorded model and reasoning effort in session viewers such as my Agent Sessions . The attached screenshot shows gpt-5.6-luna , thinking max , and the change from xhigh to max in its Session Info history.