OpenAI's Codex 0.133.0 adds default goals, remote control, and permission profiles

OpenAI has released Codex 0.133.0, a significant update to its AI-powered coding assistant. The headline feature is that Goals are now enabled by default, backed by dedicated storage, and can track progress across active turns. This allows developers to set persistent objectives that Codex works toward over multiple interactions. The remote control CLI has been overhauled: the new "codex remote-control" command runs like a foreground process, waits for readiness, reports machine status, and includes explicit daemon-style start/stop controls for headless environments.

Permission profiles see major enhancements, including list APIs, inheritance, managed requirements.toml support, runtime refresh behavior, and stronger Windows sandbox integration. Plugin discovery is now more inspectable, with marketplace-aware list output showing installed versions and remote collection support. Extensions gain access to additional lifecycle events, such as subagent start/stop, tool execution, turn metadata, and async approval processing. Bug fixes address TUI working directory issues, plan-mode enter key handling, stale background poll events, and app-server startup/shutdown races. The release also improves release packaging with a canonical archive pipeline and fixes Linux Python runtime wheel tags for glibc-based systems.