Azure custom provider: model refresh logs "missing field `models`" on every exec, and features.remote_models=false does not suppress it

Version

codex-cli 0.130.0

Config

Using an Azure custom provider in ~/.codex/config.toml:

model = "gpt-5.4"
model_provider = "azure"

[model_providers.azure]
name = "Azure"
base_url = "https://<redacted>.openai.azure.com/openai"
env_key = "AZURE_OPENAI_API_KEY"
wire_api = "responses"
query_params = { "api-version" = "2025-04-01-preview" }

Repro

SCRATCH=$(mktemp -d)
cd "$SCRATCH"
git init -q
printf 'test\n' > README.md

codex exec --sandbox read-only -p gpt-54-medium --ephemeral "Reply with exactly: SMOKE_OK" --skip-git-repo-check

Also reproducible with:

codex --config features.remote_models=false exec --sandbox read-only -p gpt-54-medium --ephemeral "Reply with exactly: SMOKE_OK" --skip-git-repo-check

Actual behavior

Every run logs this error before the command completes successfully:

ERROR codex_models_manager::manager: failed to refresh available models: stream disconnected before completion: failed to decode models response: missing field `models`; body: {
  "data": [
    ...
  ],
  "object": "list"
}

The run itself succeeds and returns the expected result:

SMOKE_OK

Expected behavior

One of these should happen:

1. Azure model refresh should accept the OpenAI-style /models response shape:

``json
{ "data": [...], "object": "list" }
``

1. If remote model fetching is unsupported for this Azure path, Codex should skip this refresh cleanly instead of logging an error every run.

1. features.remote_models=false should suppress this refresh path if that is the intended control.

Notes

This looks related to Azure / remote model fetching behavior in #9536.

The confusing part as a user is:

• codex exec works
• but model refresh logs a hard error every run
• and features.remote_models=false does not prevent it