Changelog UpdateOpenAI/Codex CLICLIWindows correctly prompts before unsafe commands when running with a read-only sandbox policy. (#9117)