call_id exceeds 64-character limit when using non-OpenAI backends (Azure AI Foundry / LiteLLM)

What version of Codex is running?

codex-cli 0.113.0

What subscription do you have?

BYOM

Which model were you using?

llm-gateway/gpt-5.4 (via LiteLLM proxy routing to Azure AI Foundry)

What platform is your computer?

macOS (darwin 25.3.0)

What terminal emulator and version are you using (if applicable)?

zsh

What issue are you seeing?

After a few tool-call round-trips, the call_id field sent in the request payload exceeds the 64-character maximum enforced by Azure AI Foundry, causing a 400 error:

litellm.BadRequestError: AzureException BadRequestError - {
  "error": {
    "message": "Invalid 'input[4].call_id': string too long. Expected a string with maximum length 64, but got a string with length 1324 instead.",
    "type": "invalid_request_error",
    "param": "input[4].call_id",
    "code": "string_above_max_length"
  }
}

Once this error occurs, the session is unrecoverable — every subsequent request in the same session fails with the same error. The only workaround is starting a new session.

Related but distinct from:

• #13270 (arguments string too long — different field, different limit)
• #10506 (encrypted_content too long — different field)
• #7094 (empty function name on Azure — fixed)

What steps can reproduce the bug?

1. Configure Codex with a LiteLLM proxy routing to Azure AI Foundry:

[model_providers.llm-gateway]
name = "llm-gateway"
base_url = "<litellm-proxy-url>"
env_key = "LLM_GATEWAY_KEY"
wire_api = "responses"

1. Start a session with a model routed through Azure AI Foundry (e.g. llm-gateway/gpt-5.4)
2. Interact until tool calls are generated (e.g. ask it to read a file or run a command)
3. After several tool-call round-trips the request is rejected with the error above

What is the expected behavior?

call_id values in the request payload should stay within the 64-character limit accepted by OpenAI-compatible backends.