Bug: Codex TUI does not render assistant messages for Azure Chat Completions (APIM), despite successful SSE and output

What version of Codex is running?

codex-cli 0.20.0

Which model were you using?

gpt-5

What platform is your computer?

Linux 6.6.87.1-microsoft-standard-WSL2 x86_64 x86_64

What steps can reproduce the bug?

Bug: Codex TUI does not render assistant messages for Azure Chat Completions (APIM), despite successful SSE and output

Scope: Codex CLI v0.20 TUI
Symptom: The TUI accepts input but renders no assistant reply. Logs show a successful streamed completion and an Output item being emitted. This appears to be a render-only defect.

Environment

• Codex CLI: v0.20 (x86\_64 musl build)
• OS: Linux x86\_64
• Terminal: standard POSIX terminal (e.g., xterm-256color)
• Auth: Entra ID (client credentials) → JWT in Authorization: Bearer …
• Backend: Azure OpenAI via APIM (exact host/deployment redacted)

Minimal configuration (redacted)

# ~/.codex/config.toml
projects = { "/path/to/project" = { trust_level = "trusted" } }

[model_providers.your-org]
name                 = "Azure"
base_url             = "<APIM_BASE>/openai/deployments/<DEPLOYMENT>"
wire_api             = "chat"
requires_openai_auth = false
env_key              = "AZURE_OPENAI_API_KEY"
# ensure Bearer header
env_http_headers     = { "Authorization" = "Bearer ${AZURE_OPENAI_API_KEY}" }
query_params         = { "api-version" = "<API_VERSION>" }

[profiles.az-qa]
model_provider = "your-org"
model          = "gpt-5"

Reproduction steps

1. Obtain a valid Entra ID access token with the correct audience and set:

``bash
export AZURE_OPENAI_API_KEY="<JWT>"
``

1. Run Codex TUI:

``bash
codex --profile az-qa -m gpt-5 "say hello"
``

1. Observe: the TUI shows no assistant message after submitting the prompt (no error displayed).

Expected behavior

• TUI should render streamed tokens and the final assistant message in the chat pane.

Actual behavior

• TUI does not display any assistant content. However, logs confirm successful request/response flow and an emitted output message.

Evidence (sanitized logs)

DEBUG Configuring session: model=gpt-5; provider=ModelProviderInfo {
  name: "Azure",
  base_url: Some("<APIM_BASE>/openai/deployments/<DEPLOYMENT>"),
  wire_api: Chat,
  query_params: Some({"api-version": "<API_VERSION>"}),
  requires_openai_auth: false,
  env_http_headers: Some({"Authorization": "Bearer ${AZURE_OPENAI_API_KEY}"})
}

DEBUG POST to <APIM_BASE>/openai/deployments/<DEPLOYMENT>/chat/completions?api-version=<API_VERSION>: {
  "model":"gpt-5",
  "messages":[ ... ],
  "stream":true,
  "tools":[ ... ]
}

# (SSE chunks omitted for brevity)

DEBUG Output item item=Message {
  role: "assistant",
  content: [OutputText { text: "Hello! How can I help today?" }]
}
DEBUG Turn completed

Notes

• Control tests succeed:

• curl to the same deploymented URL → 200 OK
• AzureOpenAI Python SDK → 200 OK
• Codex transport/auth are fine; only the TUI fails to paint the already-produced assistant output.
• No OpenAI key/env fallback involved (requires_openai_auth=false, explicit Bearer header).

Requested fix / investigation

• TUI should render the assistant message when an Output item is emitted for Azure Chat Completions (streaming).
• Suspect: alt-screen repaint or post-stream flush path not triggered for the Azure provider route, even though the message is delivered to the output pipeline.

Additional info

• If needed, I can provide a full RUST_LOG=trace excerpt privately with host/deployment redacted.
• I can also test a patched build to confirm the repaint path.

---

See also https://github.com/openai/codex/issues/2025 and https://github.com/openai/codex/issues/2024

What is the expected behavior?

_No response_

What do you see instead?

_No response_

Additional information

_No response_