"""Agent执行逻辑单元测试 - ContentGeneratorAgent / DeAIAgent / GEOOptimizerAgent 测试策略: - 使用 FakeLLMProvider mock LLM 调用，避免真实网络请求 - patch BaseAgent.report_progress 避免 Redis / 数据库依赖 - patch RAGService / AsyncSessionLocal 避免真实数据库访问 """ import json import uuid from datetime import datetime, timezone from unittest.mock import AsyncMock, MagicMock, patch import pytest from app.agent_framework.agents.content_generator_agent import ContentGeneratorAgent from app.agent_framework.agents.deai_agent import DeAIAgent from app.agent_framework.agents.geo_optimizer_agent import GEOOptimizerAgent from app.agent_framework.protocol import TaskMessage from app.services.llm import LLMProvider, LLMResponse, LLMError # --------------------------------------------------------------------------- # FakeLLMProvider - 测试用假LLM # --------------------------------------------------------------------------- class FakeLLMProvider(LLMProvider): """测试用假LLM，返回预设响应""" def __init__(self, response_content: str = "fake response"): self._response = response_content @property def provider_name(self) -> str: return "fake" @property def model_name(self) -> str: return "fake-model" @property def max_context_length(self) -> int: return 4096 async def chat(self, messages, **kwargs) -> LLMResponse: return LLMResponse( content=self._response, model="fake-model", usage={"prompt_tokens": 10, "completion_tokens": 20, "total_tokens": 30}, ) async def chat_stream(self, messages, **kwargs): for word in self._response.split(): yield word + " " # --------------------------------------------------------------------------- # 测试辅助函数 # --------------------------------------------------------------------------- def _make_task(task_type: str, input_data: dict) -> TaskMessage: return TaskMessage( task_id=str(uuid.uuid4()), agent_name="test_agent", task_type=task_type, priority=1, input_data=input_data, callback_url=None, created_at=datetime.now(timezone.utc), timeout_seconds=300, ) # --------------------------------------------------------------------------- # ContentGeneratorAgent 测试 # --------------------------------------------------------------------------- @pytest.mark.asyncio async def test_generate_topics_returns_parsed_json(): """FakeLLM返回JSON数组，验证topics字段正确解析""" agent = ContentGeneratorAgent() fake_llm = FakeLLMProvider( response_content='[{"title": "AI营销趋势", "reason": "热门话题"}]' ) with patch.object(agent, "report_progress", new_callable=AsyncMock): with patch( "app.agent_framework.agents.content_generator_agent.LLMFactory.get_default", return_value=fake_llm, ): task = _make_task("generate_topics", {"target_keyword": "AI营销"}) result = await agent.execute(task) assert result.status == "completed" assert "topics" in result.output_data topics = result.output_data["topics"] assert isinstance(topics, list) assert topics[0]["title"] == "AI营销趋势" assert topics[0]["reason"] == "热门话题" @pytest.mark.asyncio async def test_generate_article_success(): """验证返回content字段""" agent = ContentGeneratorAgent() fake_llm = FakeLLMProvider(response_content="这是一篇测试文章") with patch.object(agent, "report_progress", new_callable=AsyncMock): with patch( "app.agent_framework.agents.content_generator_agent.LLMFactory.get_default", return_value=fake_llm, ): task = _make_task("generate_article", {"target_keyword": "AI营销"}) result = await agent.execute(task) assert result.status == "completed" assert result.output_data["content"] == "这是一篇测试文章" assert result.output_data["word_count"] == len("这是一篇测试文章") assert "usage" in result.output_data @pytest.mark.asyncio async def test_generate_with_rag_context(): """Mock RAGService，验证知识上下文被注入""" agent = ContentGeneratorAgent() fake_llm = FakeLLMProvider( response_content='[{"title": "RAG测试选题", "reason": "测试"}]' ) # Mock AsyncSessionLocal 上下文管理器 mock_session = AsyncMock() mock_session.__aenter__ = AsyncMock(return_value=mock_session) mock_session.__aexit__ = AsyncMock(return_value=False) mock_local = MagicMock() mock_local.return_value.__aenter__ = AsyncMock(return_value=mock_session) mock_local.return_value.__aexit__ = AsyncMock(return_value=False) with patch.object(agent, "report_progress", new_callable=AsyncMock): with patch("app.database.AsyncSessionLocal", mock_local): with patch( "app.services.knowledge.rag_service.RAGService" ) as MockRAG: mock_rag = MockRAG.return_value mock_rag.search = AsyncMock( return_value=[ {"document_title": "知识库文档", "content": "相关知识内容"} ] ) with patch( "app.agent_framework.agents.content_generator_agent.LLMFactory.get_default", return_value=fake_llm, ): task = _make_task( "generate_topics", {"target_keyword": "AI营销", "knowledge_base_ids": ["kb-1"]}, ) result = await agent.execute(task) assert result.status == "completed" mock_rag.search.assert_awaited_once() # 验证 search 调用参数 call_kwargs = mock_rag.search.call_args.kwargs assert call_kwargs["query"] == "AI营销" assert call_kwargs["knowledge_base_ids"] == ["kb-1"] @pytest.mark.asyncio async def test_llm_error_returns_failed(): """Mock LLM抛出LLMError，验证返回failed状态""" agent = ContentGeneratorAgent() class ErrorLLM(FakeLLMProvider): async def chat(self, messages, **kwargs) -> LLMResponse: raise LLMError("API错误", provider="fake", status_code=500) error_llm = ErrorLLM() with patch.object(agent, "report_progress", new_callable=AsyncMock): with patch( "app.agent_framework.agents.content_generator_agent.LLMFactory.get_default", return_value=error_llm, ): task = _make_task("generate_topics", {"target_keyword": "AI营销"}) result = await agent.execute(task) assert result.status == "failed" assert "LLM调用失败" in result.error_message @pytest.mark.asyncio async def test_extract_json_from_code_block(): """测试```json```包裹的JSON提取""" agent = ContentGeneratorAgent() text = '```json\n[{"title": "测试"}]\n```' result = agent._extract_json(text) assert result == '[{"title": "测试"}]' # --------------------------------------------------------------------------- # DeAIAgent 测试 # --------------------------------------------------------------------------- @pytest.mark.asyncio async def test_deai_success(): """正常处理返回success""" agent = DeAIAgent() fake_llm = FakeLLMProvider(response_content="去AI化后的内容") with patch.object(agent, "report_progress", new_callable=AsyncMock): with patch( "app.agent_framework.agents.deai_agent.LLMFactory.get_default", return_value=fake_llm, ): task = _make_task("deai_process", {"content": "原始的AI生成内容"}) result = await agent.execute(task) assert result.status == "completed" assert result.output_data["content"] == "去AI化后的内容" assert result.output_data["original_word_count"] == len("原始的AI生成内容") assert result.output_data["processed_word_count"] == len("去AI化后的内容") @pytest.mark.asyncio async def test_deai_empty_content_fails(): """空content返回failed""" agent = DeAIAgent() fake_llm = FakeLLMProvider(response_content="something") with patch.object(agent, "report_progress", new_callable=AsyncMock): with patch( "app.agent_framework.agents.deai_agent.LLMFactory.get_default", return_value=fake_llm, ): task = _make_task("deai_process", {"content": ""}) result = await agent.execute(task) assert result.status == "failed" # ValueError 会被外层 except 捕获，error_message 包含原始异常信息 assert "content" in result.error_message.lower() or "input_data" in result.error_message.lower() @pytest.mark.asyncio async def test_deai_temperature_is_high(): """验证调用LLM时temperature=0.9""" agent = DeAIAgent() mock_provider = AsyncMock() mock_provider.chat = AsyncMock( return_value=LLMResponse( content="processed", model="fake", usage={"prompt_tokens": 10, "completion_tokens": 20, "total_tokens": 30}, ) ) with patch.object(agent, "report_progress", new_callable=AsyncMock): with patch( "app.agent_framework.agents.deai_agent.LLMFactory.get_default", return_value=mock_provider, ): task = _make_task("deai_process", {"content": "some content"}) await agent.execute(task) mock_provider.chat.assert_awaited_once() _, kwargs = mock_provider.chat.call_args assert kwargs.get("temperature") == 0.9 # --------------------------------------------------------------------------- # GEOOptimizerAgent 测试 # --------------------------------------------------------------------------- @pytest.mark.asyncio async def test_geo_optimize_json_response(): """FakeLLM返回标准JSON，验证解析""" agent = GEOOptimizerAgent() fake_llm = FakeLLMProvider( response_content=json.dumps( { "optimized_content": "优化后的文章", "seo_score": 85, "changes": ["优化了标题"], } ) ) with patch.object(agent, "report_progress", new_callable=AsyncMock): with patch( "app.agent_framework.agents.geo_optimizer_agent.LLMFactory.get_default", return_value=fake_llm, ): task = _make_task( "geo_optimize", {"content": "原始文章", "target_keywords": ["SEO"]}, ) result = await agent.execute(task) assert result.status == "completed" assert result.output_data["optimized_content"] == "优化后的文章" assert result.output_data["seo_score"] == 85 assert result.output_data["changes"] == ["优化了标题"] assert "usage" in result.output_data @pytest.mark.asyncio async def test_geo_optimize_fallback(): """FakeLLM返回纯文本，验证降级处理""" agent = GEOOptimizerAgent() fake_llm = FakeLLMProvider(response_content="这不是JSON格式") with patch.object(agent, "report_progress", new_callable=AsyncMock): with patch( "app.agent_framework.agents.geo_optimizer_agent.LLMFactory.get_default", return_value=fake_llm, ): task = _make_task( "geo_optimize", {"content": "原始文章", "target_keywords": ["SEO"]}, ) result = await agent.execute(task) assert result.status == "completed" assert result.output_data["optimized_content"] == "这不是JSON格式" assert result.output_data["seo_score"] is None assert result.output_data["changes"] == ["LLM输出非标准格式，已返回原始优化结果"] @pytest.mark.asyncio async def test_geo_optimize_keywords_in_prompt(): """验证关键词出现在渲染后的prompt variables中""" agent = GEOOptimizerAgent() with patch.object(agent, "report_progress", new_callable=AsyncMock): with patch( "app.agent_framework.agents.geo_optimizer_agent.GEO_OPTIMIZER_TEMPLATE.render" ) as mock_render: mock_render.return_value = [ {"role": "system", "content": "test prompt"} ] with patch( "app.agent_framework.agents.geo_optimizer_agent.LLMFactory.get_default" ) as mock_factory: mock_provider = AsyncMock() mock_provider.chat = AsyncMock( return_value=LLMResponse( content=json.dumps({"optimized_content": "test"}), model="fake", usage={}, ) ) mock_factory.return_value = mock_provider task = _make_task( "geo_optimize", {"content": "原始文章", "target_keywords": ["SEO", "GEO优化"]}, ) await agent.execute(task) mock_render.assert_called_once() variables = mock_render.call_args[0][0] assert "SEO" in variables["target_keywords"] assert "GEO优化" in variables["target_keywords"]