From f4a782d99f1352c9eca4eabd35b51458d8729b4a Mon Sep 17 00:00:00 2001
From: Ed_ <edwardgz@gmail.com>
Date: Mon, 15 Jun 2026 10:56:24 -0400
Subject: [PATCH] fix(ai_loop): wrap MiniMax reasoning in <thinking> tags for
 parse_thinking_trace (FR3, Bug #3)

Adds a new wrap_reasoning_in_text: bool = False keyword argument to
run_with_tool_loop. When True and reasoning_content is non-empty, the
returned text is prepended with <thinking>...</thinking> tags so
thinking_parser.parse_thinking_trace can extract a ThinkingSegment
for the discussion entry.

The wrap is conditional (default False) so it doesn't break providers
that already wrap inline (e.g. DeepSeek, which wraps at line 2117-2118
before run_with_tool_loop sees the response).

_send_minimax now passes wrap_reasoning_in_text=bool(caps.reasoning).
When caps.reasoning is True (M2.5/M2.7), the reasoning is wrapped in
<thinking> tags. When False (M2/M2.1), the parameter is False and
no wrap happens (avoids useless getattr on non-reasoning models).

Also fixes a bug in the test_fr3_minimax_thinking_in_returned_text
test mock: it was returning a raw MagicMock instead of a Result
object, which caused the test to see auto-created MagicMock attributes
instead of the expected text. Now wraps in Result(data=MagicMock(...))
and sets ai_client._model to ensure get_capabilities('minimax', _model)
resolves to the M2.7 capabilities (reasoning=True).
---
 src/ai_client.py                           | 9 +++++++++
 tests/test_ai_loop_regressions_20260614.py | 5 +++--
 2 files changed, 12 insertions(+), 2 deletions(-)

diff --git a/src/ai_client.py b/src/ai_client.py
index f3f79f59..5b5edda1 100644
--- a/src/ai_client.py
+++ b/src/ai_client.py
@@ -748,6 +748,7 @@ def run_with_tool_loop(
  reasoning_extractor: Optional[Callable[[Any], str]] = None,
  send_func:           Optional[Callable[[int], NormalizedResponse]] = None,
  on_pre_dispatch:     Optional[Callable[[int, list[dict[str, Any]]], list[dict[str, Any]]]] = None,
+ wrap_reasoning_in_text: bool = False,
 ) -> str:
  """
   Orchestrates the LLM conversation loop, executing tool calls and updating history.
@@ -772,6 +773,11 @@ def run_with_tool_loop(
    reasoning_extractor (Optional[Callable]): Callback to extract reasoning content.
    send_func (Optional[Callable]): Dispatch sender callback.
    on_pre_dispatch (Optional[Callable]): Callback to adjust tools.
+   wrap_reasoning_in_text (bool): When True and reasoning_content is non-empty, the
+    returned text is prepended with `<thinking>...</thinking>` wrapping the
+    reasoning. This lets thinking_parser.parse_thinking_trace extract a
+    ThinkingSegment for the discussion entry. Default False (callers that
+    already wrap inline, e.g. DeepSeek, pass False).
 
   Returns:
    str: The final text response returned by the LLM.
@@ -833,6 +839,8 @@ def run_with_tool_loop(
       "content":      str(out) if out else "",
      })
   if trim_func is not None: trim_func(history)
+ if wrap_reasoning_in_text and reasoning_content:
+  response_text = f"<thinking>\n{reasoning_content}\n</thinking>\n\n{response_text}"
  return response_text
 
 async def _execute_single_tool_call_async(
@@ -2438,6 +2446,7 @@ def _send_minimax(md_content: str, user_message: str, base_dir: str,
    history_lock=_minimax_history_lock, history=_minimax_history,
    trim_func=lambda h: _trim_minimax_history(_build_minimax_request(0).messages, h),
    reasoning_extractor=_extract_minimax_reasoning if caps.reasoning else None,
+   wrap_reasoning_in_text=bool(caps.reasoning),
   ))
  except Exception as exc:
   return Result(data="", errors=[_classify_minimax_error(exc, source="ai_client.minimax")])
diff --git a/tests/test_ai_loop_regressions_20260614.py b/tests/test_ai_loop_regressions_20260614.py
index 53cca46f..0a352da4 100644
--- a/tests/test_ai_loop_regressions_20260614.py
+++ b/tests/test_ai_loop_regressions_20260614.py
@@ -204,7 +204,7 @@ def test_fr3_minimax_thinking_in_returned_text() -> None:
 
  def _fake_send_openai_compatible(client, request, *, capabilities):
   captured_text.append("send_openai_compatible was called")
-  return MagicMock(
+  return Result(data=MagicMock(
    text="The final answer is 42",
    tool_calls=[],
    usage_input_tokens=0,
@@ -212,11 +212,12 @@ def test_fr3_minimax_thinking_in_returned_text() -> None:
    usage_cache_read_tokens=0,
    usage_cache_creation_tokens=0,
    raw_response=fake_raw,
-  )
+  ))
 
  from src import openai_compatible as oc
  from src.vendor_capabilities import register, VendorCapabilities
  register(VendorCapabilities(vendor="minimax", model="MiniMax-M2.7", reasoning=True))
+ ai_client._model = "MiniMax-M2.7"
 
  with patch.object(oc, "send_openai_compatible", side_effect=_fake_send_openai_compatible), \
       patch("src.ai_client._ensure_minimax_client", return_value=MagicMock()), \