Make gpt-5.5 not get distracted by excessive "Frontend guidance" in system prompt

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

Version 26.422.30944 (2080)

What subscription do you have?

Business

What platform is your computer?

macOS

What issue are you seeing?

The system prompt in codex gpt-5.5 appears to contain ~1.5K tokens specialized on building graphical frontends [1]. A large amount of use cases involve only backend work. Forcing these excessive "Frontend guidance" instructions onto every user amounts to systematic context pollution and makes the model less effective (and unnecessarily expensive) for pure backend work or analytic work.

Please make codex more effective rather than less effective.

If progressive discovery works so well, why not move those instructions into a skill or similar? Or have a config knob in config.toml to enable/disable these frontend instructions? Or find another solution that doesn't penalize the rest of the world.

BTW, codex gpt-5.5 measured loading ~15K (!) input tokens for a simple "hi" prompt with a one sentence reply, in a fresh session, with zero MCPs, zero skills enabled, zero plugins enabled, memories disabled, and single-line AGENTS.md. I know this is cached. But still, if injecting such large system/tool/permission blurbs is indeed best practice then it indicates that most guidance/tools can only be effective when loaded directly upfront, progressive discovery/skills does not work well, RL didn't work well, and pretraining intelligence and general intelligence is largely absent.

[1] https://github.com/asgeirtj/system_prompts_leaks/blob/86a95b27da09282888cd9ccb7779cbd15c455658/OpenAI/codex/gpt-5.5.md

What steps can reproduce the bug?

Type something like "Helllo" in a fresh session.

What is the expected behavior?

System instructions that are helpful rather than distracting.

Additional information

_No response_