Imagine you are asking a coding assistant to finish a job, not just answer one prompt. OpenAI’s Codex Goal mode is a feature that lets a user define an outcome and success criteria, then have Codex continue working toward that objective. It matters because many software tasks are not single-turn requests. They involve checking files, making changes, running validation, and deciding whether the stated end condition has been reached.
Goal mode is for Codex users who need a clearer way to manage longer coding work. OpenAI describes Codex as a coding agent that helps users build and ship with AI, including work such as features, refactors, migrations, reviews, and releases. The feature is especially relevant to developers and teams handling long-running coding work with a clear success condition and validation loop. Public sources identify code migrations, large refactors, deployment retry loops, experiments, games, side projects, and long experiments with clear success criteria as suitable uses.
Goal mode fits inside the Codex app, the IDE extension, and the CLI. OpenAI’s ChatGPT release notes say the feature became generally available across those surfaces on May 21, 2026. It is most useful when a task needs Codex to work across turns toward a verifiable stopping condition. OpenAI’s guidance says a good goal should be bigger than one prompt but smaller than an open-ended backlog. Like giving a courier both a destination and proof of delivery, the feature works best when the endpoint is concrete.
In practice, the user starts Goal mode with the slash command “/goal” and supplies an objective. OpenAI says the goal text acts as both the starting prompt and the completion criteria, helping Codex decide what to do next and whether the task is complete. Users can set a goal, check the current goal, pause it, resume it, or clear it. In the Codex app, progress appears above the composer with controls to pause, resume, edit, or clear the goal. If the slash command does not appear, OpenAI’s documentation says users can enable “features.goals” in the Codex configuration file or run a CLI command to enable goals.
The next step is practical rather than abstract: define one coding objective with a measurable finish line before using Goal mode. OpenAI’s examples emphasize specific outcomes, target metrics, tests, or other criteria that prove progress. A team considering the feature should avoid loose lists of unrelated work and instead state what Codex should achieve, what it should not change, how it should validate progress, and when it should stop. Public sources do not clearly confirm every possible limitation of Goal mode, so users should rely on OpenAI’s current documentation and review Codex’s work through normal engineering practices such as tests, diffs, and version control.