Codex reverts changes made to `constant values` during a second prompt / task

What version of Codex is running?

0.55.0

What subscription do you have?

Pro

Which model were you using?

gpt-5-codex-high

What platform is your computer?

macOS

What issue are you seeing?

• Ask Codex to perform some work such as adding animation to some existing code
• After it's done, manually change the animation timing to something else, e.g. if it chose 0.35 seconds, change it to 0.25 seconds.
• Now ask Codex to add a new improvement - it notices the change to 0.25 and _reverts_ it back to 0.35! This is a trivial example but many times it's more than this.

This is related to https://github.com/openai/codex/issues/6187 but not entirely. This is where it's performing _new_ work but insists on keeping the code exactly how it was when it finished working on it. Any deviation and it thinks this is unwanted and changes it back. Often you wan manual and complete control over the final code but the only solution is to always /new to avoid this.

What steps can reproduce the bug?

See above.

What is the expected behavior?

_No response_

Additional information

_No response_