fix: treat SSE stream without [DONE] as truncated (retry, no cache)

adapters/llm.py

@@ -578,6 +578,10 @@ class GovernedLLMClient:
             else:
                 thinking_parts.append(text)
 
+        # 流耗尽但未收 [DONE] → 服务端截断，视为可重试的 SSE 异常（不写缓存/不当成功）
+        if not usage_sink.get("done"):
+            raise _SseAnomaly("truncated_no_done")
+
         content = "".join(content_parts)
         thinking = "".join(thinking_parts)
         usage = usage_sink.get("usage", {})

tests/unit/test_governed_llm.py

@@ -259,6 +259,33 @@ async def test_qwen_thinking_stripped():
     assert thinking2 == ""
 
 
+@pytest.mark.asyncio
+async def test_truncated_stream_without_done_raises():
+    """SSE 流耗尽但未收 [DONE] → _SseAnomaly（进重试，不当成功）。"""
+    from adapters.llm import _SseAnomaly
+
+    async def _lines():
+        yield 'data: {"choices":[{"delta":{"content":"半"}}]}'
+        # 无 data: [DONE] —— 模拟服务端截断
+
+    client = _build_client()
+    with pytest.raises(_SseAnomaly):
+        await client._consume_stream(_lines())
+
+
+@pytest.mark.asyncio
+async def test_complete_stream_with_done_ok():
+    """正常带 [DONE] 的流不抛异常，正确累积 content。"""
+
+    async def _lines():
+        yield 'data: {"choices":[{"delta":{"content":"完整"}}]}'
+        yield "data: [DONE]"
+
+    client = _build_client()
+    content, _thinking, _ttft, _gap, _usage = await client._consume_stream(_lines())
+    assert content == "完整"
+
+
 @pytest.mark.asyncio
 async def test_parent_call_id_forwarded_to_telemetry():
     """parent_call_id 和 session_id 正确传递到遥测记录。"""