Feature request: Support model-scoped configuration overrides

Summary

It would be useful if Codex could automatically apply configuration overrides based on the currently selected model.

Different models can benefit from different harness configurations. For example, a highly agentic model may work better with a minimal harness, while another model may benefit from plugins, apps, tool suggestions, or other Codex features.

Currently, switching models with /model changes the model, but does not provide a general mechanism for changing the associated Codex configuration at the same time.

Example

Something along these lines:

[model_overrides."gpt-6-astra".features]
plugins = false
apps = false
tool_suggest = false

[model_overrides."gpt-5.6-sol".features]
plugins = true
apps = true
tool_suggest = true

Then:

/model -> gpt-6-astra
             |
             v
      model-scoped config
             |
             v
      plugins = false
      apps = false
      tool_suggest = false
             |
             v
       effective config

The exact TOML structure is only an example; I am not attached to this particular syntax.

Why

There already appear to be cases where configuration is effectively model-dependent.

Related issues include:

• #37010 — configuring remote_compaction per model
• #17261 — per-model context-window configuration
• #22160 — exposing profiles / provider-model aliases through the model picker

Rather than introducing separate per-model configuration mechanisms for individual options, it may be useful to have a general model-scoped configuration layer that composes with the existing configuration system.

This could also be useful for model/tool compatibility. A model may support or perform well with one set of Codex features while another model may require or benefit from a different set.

Configuration precedence

If implemented as a general configuration layer, the precedence should probably be explicitly defined.

For example:

base/user config
      |
      v
model-scoped overrides
      |
      v
profile/project config
      |
      v
CLI/session overrides
      |
      v
effective config

The exact precedence is open for discussion; the important property would be deterministic composition with the existing configuration system.

AI disclosure

This issue is intentionally being created with substantial assistance from GPT.

GPT was used to:

1. investigate the current Codex configuration/features behavior;
2. search existing Codex issues for similar proposals;
3. compare related issues and identify what appears to be a missing general case;
4. help formulate this feature request; and
5. draft and submit this issue.

The decision to submit the feature request and the underlying use case are mine, but both the research and the issue text were produced with GPT assistance.

I am explicitly disclosing this because the subject is Codex/agent behavior itself, and I think it is useful context for maintainers reviewing the proposal.

If this direction fits the intended Codex configuration architecture, I am also interested in implementing it and submitting a PR.

No files or tests are named. Start by tracing the /model selection flow and existing configuration or profile precedence, then compare related issues #37010, #17261, and #22160. Done means an agreed model-scoped configuration design with explicit precedence and implementation coverage for the requested behavior.