feat：修复部分模型请求问题(v4l）

src/maisaka/chat_loop_service.py

@@ -35,7 +35,7 @@ from .context_messages import (
     ToolResultMessage,
     build_llm_message_from_context,
 )
-from .history_utils import drop_orphan_tool_results
+from .history_utils import drop_orphan_tool_results, normalize_tool_result_order
 from .display.prompt_cli_renderer import PromptCLIVisualizer
 from .visual_mode_utils import resolve_enable_visual_planner
 
@@ -652,6 +652,7 @@ class MaisakaChatLoopService:
         selected_history = [filtered_history[index] for index in selected_indices]
         selected_history, _ = MaisakaChatLoopService._hide_early_assistant_messages(selected_history)
         selected_history, _ = drop_orphan_tool_results(selected_history)
+        selected_history, _ = normalize_tool_result_order(selected_history)
         tool_message_count = sum(1 for message in selected_history if isinstance(message, ToolResultMessage))
         normal_message_count = len(selected_history) - tool_message_count
         selection_reason = (

src/maisaka/history_post_processor.py

@@ -3,7 +3,7 @@
 from dataclasses import dataclass
 
 from .context_messages import AssistantMessage, LLMContextMessage, ToolResultMessage
-from .history_utils import drop_leading_orphan_tool_results, drop_orphan_tool_results
+from .history_utils import drop_leading_orphan_tool_results, drop_orphan_tool_results, normalize_tool_result_order
 
 TIMING_HISTORY_TOOL_NAMES = {"continue", "finish", "no_reply", "wait"}
 EARLY_TRIM_RATIO = 0.2
@@ -15,6 +15,7 @@ class HistoryPostProcessResult:
 
     history: list[LLMContextMessage]
     removed_count: int
+    changed_count: int
     remaining_context_count: int
 
 
@@ -30,6 +31,7 @@ def process_chat_history_after_cycle(
     removed_assistant_thought_count = _remove_early_assistant_thoughts(processed_history)
 
     processed_history, orphan_removed_count = drop_orphan_tool_results(processed_history)
+    processed_history, moved_tool_result_count = normalize_tool_result_order(processed_history)
     remaining_context_count = sum(1 for message in processed_history if message.count_in_context)
     removed_overflow_count = 0
 
@@ -48,9 +50,11 @@ def process_chat_history_after_cycle(
         + orphan_removed_count
         + removed_overflow_count
     )
+    changed_count = removed_count + moved_tool_result_count
     return HistoryPostProcessResult(
         history=processed_history,
         removed_count=removed_count,
+        changed_count=changed_count,
         remaining_context_count=remaining_context_count,
     )
 

src/maisaka/history_utils.py

@@ -105,3 +105,70 @@ def drop_orphan_tool_results(
         filtered_history.append(message)
 
     return filtered_history, removed_count
+
+
+def normalize_tool_result_order(
+    chat_history: list[LLMContextMessage],
+) -> tuple[list[LLMContextMessage], int]:
+    """把被其他消息隔开的 tool 结果移动到对应 assistant tool_calls 后面。"""
+
+    if not chat_history:
+        return chat_history, 0
+
+    consumed_indexes: set[int] = set()
+    normalized_history: list[LLMContextMessage] = []
+    moved_count = 0
+
+    for index, message in enumerate(chat_history):
+        if index in consumed_indexes:
+            continue
+
+        normalized_history.append(message)
+        if not isinstance(message, AssistantMessage) or not message.tool_calls:
+            continue
+
+        appended_tool_result_count = 0
+        for tool_call in message.tool_calls:
+            tool_call_id = str(tool_call.call_id or "").strip()
+            if not tool_call_id:
+                continue
+
+            matching_index = _find_tool_result_index(
+                chat_history,
+                tool_call_id=tool_call_id,
+                start_index=index + 1,
+                consumed_indexes=consumed_indexes,
+            )
+            if matching_index is None:
+                continue
+
+            consumed_indexes.add(matching_index)
+            normalized_history.append(chat_history[matching_index])
+            expected_index = index + appended_tool_result_count + 1
+            if matching_index != expected_index:
+                moved_count += 1
+            appended_tool_result_count += 1
+
+    if moved_count <= 0:
+        return chat_history, 0
+    return normalized_history, moved_count
+
+
+def _find_tool_result_index(
+    chat_history: list[LLMContextMessage],
+    *,
+    tool_call_id: str,
+    start_index: int,
+    consumed_indexes: set[int],
+) -> int | None:
+    """查找指定 tool_call_id 对应的 tool 结果消息位置。"""
+
+    for index in range(start_index, len(chat_history)):
+        if index in consumed_indexes:
+            continue
+        message = chat_history[index]
+        if not isinstance(message, ToolResultMessage):
+            continue
+        if message.tool_call_id == tool_call_id:
+            return index
+    return None

src/maisaka/reasoning_engine.py

@@ -341,6 +341,13 @@ class MaisakaReasoningEngine:
                 queued_trigger = await self._runtime._internal_turn_queue.get()
                 message_triggered, timeout_triggered = self._drain_ready_turn_triggers(queued_trigger)
 
+                if self._runtime._agent_state == self._runtime._STATE_WAIT and not timeout_triggered:
+                    self._runtime._message_turn_scheduled = False
+                    logger.debug(
+                        f"{self._runtime.log_prefix} 当前仍处于 wait 状态，忽略消息触发并继续等待超时"
+                    )
+                    continue
+
                 if message_triggered:
                     await self._runtime._wait_for_message_quiet_period()
                     self._runtime._message_turn_scheduled = False
@@ -809,10 +816,12 @@ class MaisakaReasoningEngine:
             self._runtime._chat_history,
             max_context_size=self._runtime._max_context_size,
         )
-        if process_result.removed_count <= 0:
+        if process_result.changed_count <= 0:
             return
 
         self._runtime._chat_history = process_result.history
+        if process_result.removed_count <= 0:
+            return
         self._runtime._log_history_trimmed(
             process_result.removed_count,
             process_result.remaining_context_count,