feat(agent): control reasoning visibility

ai_knowledge_base_v2/apps/backend/tests/test_agent_runtime_config.py

@@ -3,6 +3,7 @@ from decimal import Decimal
 import json
 from unittest.mock import patch
 
+import pytest
 from sqlalchemy import create_engine
 from sqlalchemy.orm import Session
 from sqlalchemy.pool import StaticPool
@@ -10,7 +11,7 @@ from sqlalchemy.pool import StaticPool
 from app.api.admin_settings import _debug_agent_stream, get_agent_runtime_config, save_agent_runtime_config
 from app.models import Base
 from app.models.admin import Admin
-from app.models.ai_config import ModelConfig
+from app.models.ai_config import ModelConfig, SystemConfig
 from app.schemas.admin import AgentDebugRequest, AgentRuntimeConfigSaveRequest
 from app.services.model_stream_service import (
     AsyncStreamingModelResponse,
@@ -57,7 +58,11 @@ def test_runtime_config_defaults_to_long_answer_safe_max_tokens():
     with _database() as db:
         admin = _admin()
         model = _model()
-        db.add_all([admin, model])
+        db.add_all([
+            admin,
+            model,
+            SystemConfig(id=1, config_key="show_model_reasoning", config_value="false"),
+        ])
         db.commit()
 
         result = get_agent_runtime_config(db=db, current_admin=admin)["data"]
@@ -66,6 +71,7 @@ def test_runtime_config_defaults_to_long_answer_safe_max_tokens():
         assert result["modelName"] == "正式模型"
         assert result["maxToken"] == 8192
         assert result["streamEnabled"] == 1
+        assert result["reasoningVisible"] == 0
         assert _max_output_tokens(model) == 8192
 
 
@@ -73,7 +79,11 @@ def test_saved_runtime_config_is_persisted_on_enabled_model():
     with _database() as db:
         admin = _admin()
         model = _model()
-        db.add_all([admin, model])
+        db.add_all([
+            admin,
+            model,
+            SystemConfig(id=1, config_key="show_model_reasoning", config_value="false"),
+        ])
         db.commit()
 
         result = save_agent_runtime_config(
@@ -85,6 +95,7 @@ def test_saved_runtime_config_is_persisted_on_enabled_model():
                 frequencyPenalty=0.4,
                 maxToken=12000,
                 streamEnabled=0,
+                reasoningVisible=1,
             ),
             db=db,
             current_admin=admin,
@@ -100,6 +111,8 @@ def test_saved_runtime_config_is_persisted_on_enabled_model():
         assert model.frequency_penalty == Decimal("0.40")
         assert model.stream_enabled == 0
         assert result["streamEnabled"] == 0
+        assert result["reasoningVisible"] == 1
+        assert db.query(SystemConfig).filter_by(config_key="show_model_reasoning").one().config_value == "true"
 
         payload = _openai_stream_payload(
             model,
@@ -133,7 +146,8 @@ def test_disabled_stream_returns_one_complete_chunk():
     assert chunks == [answer]
 
 
-def test_agent_debug_stream_emits_status_content_and_trace():
+@pytest.mark.parametrize("reasoning_visible", [True, False])
+def test_agent_debug_stream_respects_reasoning_visibility(reasoning_visible):
     async def chunks():
         yield "<think>内部思考</think>"
         yield "- **第一步**：停一下"
@@ -162,6 +176,11 @@ def test_agent_debug_stream_emits_status_content_and_trace():
         admin = _admin()
         model = _model()
         db.add_all([admin, model])
+        db.add(SystemConfig(
+            id=1,
+            config_key="show_model_reasoning",
+            config_value="true" if reasoning_visible else "false",
+        ))
         db.commit()
         payload = AgentDebugRequest(
             promptContent="你是测试助手",
@@ -187,10 +206,18 @@ def test_agent_debug_stream_emits_status_content_and_trace():
         for event in events
         if event != "data: [DONE]\n\n"
     ]
-    assert [event["type"] for event in decoded] == ["status", "content", "content", "complete"]
-    assert decoded[1]["content"].startswith("<think>")
-    assert decoded[2]["content"] == "- **第一步**：停一下"
-    assert decoded[3]["retrievalTrace"][0]["tool"] == "KnowledgeSearch"
+    expected_types = (
+        ["status", "reasoning", "content", "complete"]
+        if reasoning_visible
+        else ["status", "content", "complete"]
+    )
+    assert [event["type"] for event in decoded] == expected_types
+    assert decoded[0]["reasoningVisible"] is reasoning_visible
+    if reasoning_visible:
+        assert decoded[1]["content"] == "内部思考"
+    content_event = next(event for event in decoded if event["type"] == "content")
+    assert content_event["content"] == "- **第一步**：停一下"
+    assert decoded[-1]["retrievalTrace"][0]["tool"] == "KnowledgeSearch"
 
 
 def test_agent_debug_does_not_truncate_long_answer(monkeypatch):

ai_knowledge_base_v2/apps/backend/tests/test_production_readiness.py

@@ -101,6 +101,7 @@ def test_queued_chat_reports_position_and_completes(monkeypatch):
     monkeypatch.setattr(chat, "wait_for_chat_slot", wait_slot)
     monkeypatch.setattr(chat, "release_chat_slot", release_slot)
     monkeypatch.setattr(chat.ChatStreamService, "stream_answer_async", stream_answer)
+    monkeypatch.setattr(chat.ReasoningPolicyService, "is_visible", lambda _db: False)
 
     payload = type("Payload", (), {"sessionId": 1, "message": "问题"})()
     async def collect_events():

ai_knowledge_base_v2/apps/backend/tests/test_reasoning_policy.py

@@ -0,0 +1,18 @@
+from app.services.reasoning_policy_service import ReasoningPolicyService, ReasoningStreamParser
+
+
+def test_reasoning_parser_handles_tags_split_across_chunks():
+    parser = ReasoningStreamParser()
+    segments = []
+    for chunk in ["<thi", "nk>内部", "思考</th", "ink>## 回答"]:
+        segments.extend(parser.feed(chunk))
+    segments.extend(parser.finish())
+
+    assert "".join(item.content for item in segments if item.kind == "reasoning") == "内部思考"
+    assert "".join(item.content for item in segments if item.kind == "content") == "## 回答"
+
+
+def test_reasoning_is_removed_from_history_text_when_policy_is_off():
+    text = "<think>不应透出</think>\n正式回答"
+
+    assert ReasoningPolicyService.strip_reasoning(text) == "\n正式回答"