[Bug]: Codex CLI integration with MiniMax profile stuck in infinite Reconnecting loop

[crash-zwt]

Summary

When using DeepScientist with Codex CLI 0.57.0 and MiniMax provider (m27 profile), the WebUI keeps showing "Reconnecting... 1/10 ... 10/10" infinitely and cannot work properly.

Steps to reproduce

1. Install Codex CLI 0.57.0 (npm install -g @openai/codex@0.57.0)
2. Configure MiniMax provider in ~/.codex/config.toml with profile m27
3. Start DeepScientist with: ds --codex-profile m27 --no-daemon
4. Open WebUI http://127.0.0.1:20999
5. The Copilot panel shows "Reconnecting... 1/10 ... 10/10" in a loop forever

Expected behavior

WebUI should connect to Codex CLI successfully and work normally with MiniMax model.

Actual behavior

WebUI Copilot is stuck in an infinite reconnecting loop (Reconnecting... 1/10 to 10/10 repeated).
No chat response, no task execution, UI cannot be used.

Environment

• DeepScientist version: 1.5.15

• Browser: Chrome

• OS: macOS

• Codex CLI: 0.57.0

• Model Provider: MiniMax (m27 profile)

• Configuration: model = MiniMax-M2.7, requires_openai_auth = false

Logs, screenshots, or stack traces

Infinite Reconnecting... 1/10, 2/10, ..., 10/10 loop in WebUI Copilot panel.
Codex CLI works fine in terminal (codex --profile m27), but DeepScientist WebUI cannot connect.

Suspected scope

Codex CLI integration layer, WebSocket connection handler, MiniMax provider adapter

[Jackson2z]

感觉使用第三方api非常不方便

[ResearAI]

We dug into this and confirmed this was not just a WebUI reconnect problem. There were a few MiniMax-specific compatibility issues in the DeepScientist/Codex path:

• DeepScientist could still override the provider profile with an explicit model
• conflicting OPENAI_* env vars could leak into MiniMax runs even when requires_openai_auth = false
• the profile-only MiniMax config shape was not being adapted consistently enough in the DeepScientist runtime path

We have now patched this in main and verified the DeepScientist path end-to-end with Codex CLI 0.57.0 + MiniMax:

• m27 with MiniMax-M2.7
• m25 with MiniMax-M2.5

We also added a startup check so that if DeepScientist detects a MiniMax profile with the wrong Codex CLI version, it will prompt to switch to @openai/codex@0.57.0.

Sorry again for the inconvenience. If you have a chance, please try the latest main and let us know whether it resolves the reconnect loop on your side.