Problem

Multi-turn conversations with Xiaomi MiMo thinking models fail with HTTP 400:

"The reasoning_content in the thinking mode must be passed back to the API"

Hermes does not echo reasoning_content back on subsequent assistant turns for MiMo, even though the same requirement is already handled for DeepSeek and Kimi.

Root cause

_needs_thinking_reasoning_pad() at run_agent.py:10095 only covered two providers:

def _needs_thinking_reasoning_pad(self) -> bool:
    return (
        self._needs_deepseek_tool_reasoning()
        or self._needs_kimi_tool_reasoning()
    )

MiMo (xiaomimimo.com, model names containing mimo) has the identical contract but was absent from this check, so _copy_reasoning_content_for_api never injected the required reasoning_content placeholder on assistant turns during history replay.

Fix

Add _needs_mimo_tool_reasoning() detecting xiaomimimo.com base URLs or model names containing "mimo", and include it in _needs_thinking_reasoning_pad():

def _needs_mimo_tool_reasoning(self) -> bool:
    model = (self.model or "").lower()
    return (
        base_url_host_matches(self.base_url, "xiaomimimo.com")
        or "mimo" in model
    )

def _needs_thinking_reasoning_pad(self) -> bool:
    return (
        self._needs_deepseek_tool_reasoning()
        or self._needs_kimi_tool_reasoning()
        or self._needs_mimo_tool_reasoning()   # ← new
    )

Tests

tests/run_agent/test_mimo_reasoning_content.py (new file):

Stash-verify: FAILED (AttributeError: 'AIAgent' object has no attribute '_needs_mimo_tool_reasoning') before fix, 3 passed after.

Visual

graph TD
    A[_needs_thinking_reasoning_pad] --> B[_needs_deepseek_tool_reasoning]
    A --> C[_needs_kimi_tool_reasoning]
    A --> D[_needs_mimo_tool_reasoning]
    D --> E{xiaomimimo.com in base_url?}
    D --> F{mimo in model name?}

Closes #24443