app-server injects client's requested personality even if disabled or overridden in config.toml

What version of the Codex App are you using (From “About Codex” dialog)?

v0.116.0

What subscription do you have?

free / ChatGPT

What platform is your computer?

Linux 6.18.12+deb13-amd64 x86_64

What issue are you seeing?

Looking at the backend-api request being sent, the "personality_friendly" is being inserted into the gpt-5.3-codex intro, even though in config.toml we have:-

[features]
personality = false

I have not yet tested if having:-

personality = "none"

Disables the client's requested personality, but the having the personality feature disabled ought to disable it, IMHO.

What steps can reproduce the bug?

Use the codex extension of Cursor connected to a v0.116 app server. (No personality is NOT an option).

What is the expected behavior?

Given the personality feature was disabled, I expected there to be no personality added.

Additional information

_No response_