codex - 💡(How to fix) Fix `Memories` in codex potentially a bad idea

What issue are you seeing?

Personal take: I suspect many users will enable this opaque feature. While it seems like an ideal feature for a chatbot, I don’t see how it can be beneficial for individuals working on different projects. For instance, I might advise against using a specific coding style for one project, but that same advice might not be applicable to another project. Since Memories are essentially Rules being applied (such as those in AGENTS.md), using the wrong Rule at the wrong time can silently lead to unexpected outcomes for the user.

I’ve noticed a recent trend of people discussing degradation, and I suspect it’s due to the activation of Memories, which are affecting GPT’s decision-making process.

What steps can reproduce the bug?

Tell codex to not do certain things in Project A. These rules may silently be memorized and applied to Project B.

What is the expected behavior?

Memories should come with an underlying warning - not suitable when you're working across projects (or different tech stacks / languages) OR for folks that change their minds often.

Additional information

No response