Persistent coding personalities maintained across model updates

What variant of Codex are you using?

App; this request also applies to CLI, IDE Extension, and Web.

What feature would you like to see?

My desired workflow changes depending on the task. Sometimes I want a persistent autonomous agent, sometimes I want pure vibe coding, and sometimes I want progress in small steps so I can check every line easily.

I've tried achieving these workflows through custom prompts, with limited success. Getting the wording right takes substantial effort, and model iterations can change how those instructions are interpreted.

Please offer built-in, selectable coding personalities that define how the agent collaborates, with behavior maintained and tested by your team across model updates.

• Persistent autonomous agent: Keep working through implementation, testing, failures, and next steps without repeatedly needing “continue.” Support open-ended work until I stop it, within configured scope and budgets.
• Vibe coding: I describe the outcome and give product feedback; the agent owns implementation choices and delivers working results with minimal supervision.
• Small-step collaboration: Make one small, coherent change at a time, explain it briefly, and pause for review. Keep each diff small enough that I can comfortably check every line before proceeding.

The core request is for your team to own that maintenance: document each personality's expected behavior, evaluate it across model releases, and adapt the underlying prompts and orchestration to preserve the user experience.

The selected personality should persist across sessions and context compaction, support project defaults, and be easy to switch during a task. It should govern initiative, stopping behavior, change size, and review cadence, alongside existing permission settings.

Success means I can select how I want to collaborate and trust that it will keep working after the next model update, without becoming a prompt engineer myself.

Additional information

Related: #11538 requests custom personalities. This request specifically asks for OpenAI-maintained workflow presets and behavioral regression testing across model updates. Merely letting users save and select their own prompts would leave the maintenance burden described above unresolved.

The issue names no files, tests, or entry points. Start by reviewing the existing custom-personality request in #11538 and comparing current behavior across the App, CLI, IDE Extension, and Web. Done would require selectable maintained workflow presets, persistence across sessions and context compaction, project defaults, switching during a task, and behavioral regression testing across model updates.