{ "metadata": { "id": "ch29", "title": "第29章：内容生成流水线", "volume": "vol8", "volume_title": "实战案例集", "word_count": 2279, "difficulty": "advanced", "prerequisites": [ "ch18" ], "key_concepts": [ "需求分析与功能规划", "业务背景", "功能清单", "非功能需求", "架构设计", "项目结构", "核心类设计", "核心代码实现", "配置与 LLM 客户端", "选题策划 Agent", "内容撰写 Agent", "审校优化 Agent", "分发适配 Agent", "敏感词过滤服务", "FastAPI 入口" ], "learning_objectives": [], "estimated_tokens": 1367, "source_file": "vol8/ch29_内容生成流水线.md" }, "overview": "", "sections": [ { "id": "29.1", "title": "29.1 需求分析与功能规划", "level": 2, "content": "", "subsections": [ { "id": "29.1.1", "title": "29.1.1 业务背景", "content": "内容营销是数字时代的核心增长引擎，但内容生产面临三大困境：\n\n1. **产能瓶颈**：一个优秀的内容创作者每周产出 3-5 篇高质量文章，远不够覆盖多渠道需求\n2. **质量不稳定**：不同创作者水平差异大，审校标准不统一，发布质量参差不齐\n3. **渠道适配难**：同一主题需要为公众号、知乎、小红书、抖音等不同平台定制不同格式和风格\n\n我们需要构建一个 AI 驱动的内容生成流水线，覆盖从选题策划到多渠道分发的全流程：\n\n- **智能选题策划**：基于热点趋势和用户画像推荐高价值选题\n- **AI 辅助写作**：基于大纲和参考资料自动生成高质量初稿\n- **自动审校优化**：语法检查、风格统一、SEO 优化、敏感词过滤\n- **多渠道适配**：一键将内容转换为不同平台的格式和风格" }, { "id": "29.1.2", "title": "29.1.2 功能清单", "content": "" }, { "id": "29.1.3", "title": "29.1.3 非功能需求", "content": "| 维度 | 指标 |\n|------|------|\n| 单篇文章生成时间 | < 30 秒（初稿） |\n| 内容原创度 | > 85%（相似度检测） |\n| 敏感词准确率 | > 95% |\n| 并发支持 | 50 QPS |\n| 支持渠道 | 微信公众号、知乎、小红书、头条号、抖音 |\n\n---" } ] }, { "id": "29.2", "title": "29.2 架构设计", "level": 2, "content": "", "subsections": [ { "id": "29.2.1", "title": "29.2.1 项目结构", "content": "" }, { "id": "29.2.2", "title": "29.2.2 核心类设计", "content": "系统由四个 Agent 组成，形成完整的内容生产 Pipeline：\n\n- **TopicAgent**：分析热点趋势和用户需求，推荐高价值选题并生成内容大纲\n- **WritingAgent**：基于大纲和参考资料撰写内容初稿，支持多种写作风格\n- **ReviewAgent**：对初稿进行多维度审校——语法、SEO、敏感词、可读性\n- **DistributeAgent**：将审校后的内容适配为不同平台的格式和风格\n\n**设计决策**：采用**流水线 + 迭代反馈**模式。ReviewAgent 发现问题后反馈给 WritingAgent 修改，最多迭代 3 次。DistributeAgent 并行适配多渠道。\n\n---" } ] }, { "id": "29.3", "title": "29.3 核心代码实现", "level": 2, "content": "", "subsections": [ { "id": "29.3.1", "title": "29.3.1 配置与 LLM 客户端", "content": "" }, { "id": "29.3.2", "title": "29.3.2 选题策划 Agent", "content": "" }, { "id": "29.3.3", "title": "29.3.3 内容撰写 Agent", "content": "" }, { "id": "29.3.4", "title": "29.3.4 审校优化 Agent", "content": "" }, { "id": "29.3.5", "title": "29.3.5 分发适配 Agent", "content": "" }, { "id": "29.3.6", "title": "29.3.6 敏感词过滤服务", "content": "" }, { "id": "29.3.7", "title": "29.3.7 FastAPI 入口", "content": "---" } ] }, { "id": "29.4", "title": "29.4 测试", "level": 2, "content": "", "subsections": [ { "id": "29.4.1", "title": "29.4.1 选题分析测试", "content": "" }, { "id": "29.4.2", "title": "29.4.2 敏感词过滤测试", "content": "" }, { "id": "29.4.3", "title": "29.4.3 可读性评分测试", "content": "---" } ] }, { "id": "29.5", "title": "29.5 部署", "level": 2, "content": "---", "subsections": [] }, { "id": "29.6", "title": "29.6 经验总结", "level": 2, "content": "", "subsections": [ { "id": "29.6.1", "title": "29.6.1 踩坑记录", "content": "**坑1：AI 生成内容的\"AI 味\"**\n\n早期生成的文章有明显的人工智能痕迹——大量\"首先...其次...最后...\"、过于工整的排比、缺乏个人观点。解决方案是**注入风格 Persona**：在 Prompt 中定义明确的写作人格（语气、偏好用词、避讳表达），并在 Few-Shot 中提供同风格范文。对于品牌内容，还需注入**品牌调性词库**。\n\n**坑2：多渠道适配的质量不一致**\n\n小红书需要大量 Emoji 和口语化表达，知乎需要学术严谨，抖音需要视觉化脚本——用同一个 Prompt 适配所有平台效果很差。我们为每个渠道编写了**专属适配 Prompt 模板**，包含平台特有的格式规范和风格指南。\n\n**坑3：审校闭环的 Token 成本**\n\n审校发现→反馈→重写→再审校的循环可能消耗大量 Token。我们将**审校分级处理**：语法和格式问题用规则引擎直接修正（不消耗 LLM），只有逻辑和风格问题才交由 LLM 处理，成本降低 60%。\n\n**坑4：敏感词误判**\n\n\"最好\"在\"最好的实践\"语境中是正常的，但在广告语境中违规。解决方案是**上下文感知过滤**：结合文案类型（广告/教程/观点）和语境判断是否为违规使用，而非简单的字符串匹配。" }, { "id": "29.6.2", "title": "29.6.2 性能优化经验", "content": "1. **并行审校**：语法检查、SEO 分析、敏感词过滤三路并行执行\n2. **分段审校**：不审校全文，而是逐段审校，发现 3 个问题后提前终止\n3. **流式输出**：文章生成使用流式输出，用户可实时看到生成进度\n4. **缓存机制**：相同选题的大纲可复用，避免重复生成" }, { "id": "29.6.3", "title": "29.6.3 关键设计模式总结", "content": "| 模式 | 应用场景 | 效果 |\n|------|---------|------|\n| Style Persona | 内容风格控制 | AI味降低 80% |\n| 规则 + LLM 混合审校 | 内容质量控制 | 成本降低 60% |\n| Pipeline + 反馈循环 | 内容质量迭代 | 一次通过率 70%→90% |\n| 渠道模板隔离 | 多渠道适配 | 各平台质量达标率 95% |" }, { "id": "29.6.4", "title": "29.6.4 未来演进方向", "content": "1. **品牌知识库**：为每个品牌构建专属风格库和术语库\n2. **A/B 测试集成**：自动生成多个标题版本，通过 A/B 测试选择最优\n3. **内容日历**：基于历史数据分析最佳发布时间和频率\n4. **多模态内容**：从图文扩展到短视频脚本、播客大纲、PPT 演讲稿\n\n---\n\n**本章小结**：内容生成流水线展示了 Agent 在内容生产领域的强大能力。核心在于**风格可控的生成**（Style Persona）和**质量闭环**（审校→反馈→重写）。通过 Pipeline 模式串联选题、撰写、审校、分发四个环节，并在每个环节注入专业规则和 LLM 能力，可以将内容生产效率提升 10 倍以上，同时保证质量的一致性。" } ] } ], "code_blocks": [ { "id": "code-1", "language": "text", "description": "- 多渠道适配：一键将内容转换为不同平台的格式和风格", "code": "┌──────────────────────────────────────────────────────────┐\n│ 内容生成流水线功能架构 │\n├──────────────────────────────────────────────────────────┤\n│ ┌─────────────┐ ┌─────────────┐ ┌──────────────┐ │\n│ │ 选题策划 │ │ 内容撰写 │ │ 审校优化 │ │\n│ │ • 热点追踪 │ │ • 大纲生成 │ │ • 语法检查 │ │\n│ │ • 竞品分析 │ │ • 段落扩写 │ │ • SEO优化 │ │\n│ │ • 关键词挖掘│ │ • 资料整合 │ │ • 敏感词过滤│ │\n│ │ • 用户画像 │ │ • 风格定制 │ │ • 可读性评分│ │\n│ └─────────────┘ └─────────────┘ └──────────────┘ │\n│ ┌─────────────────────────────────────────────────┐ │\n│ │ 多渠道分发层 │ │\n│ │ • 微信公众号 • 知乎 • 小红书 • 头条号 │ │\n│ │ • 抖音脚本 • SEO文章 • 邮件营销 │ │\n│ └─────────────────────────────────────────────────┘ │\n└──────────────────────────────────────────────────────────┘", "section_ref": "29.1.2", "runnable": false, "dependencies": [] }, { "id": "code-2", "language": "text", "description": "", "code": "content-pipeline/\n├── app/\n│ ├── main.py # FastAPI 入口\n│ ├── config.py # 配置管理\n│ ├── models/ # 数据模型\n│ │ ├── topic.py # 选题模型\n│ │ ├── article.py # 文章模型\n│ │ └── channel.py # 渠道配置\n│ ├── agents/ # Agent 核心\n│ │ ├── topic_agent.py # 选题策划 Agent\n│ │ ├── writing_agent.py # 内容撰写 Agent\n│ │ ├── review_agent.py # 审校优化 Agent\n│ │ └── distribute_agent.py # 分发适配 Agent\n│ ├── services/\n│ │ ├── plagiarism_checker.py # 原创度检测\n│ │ ├── sensitive_filter.py # 敏感词过滤\n│ │ └── seo_optimizer.py # SEO 优化\n│ └── utils/\n│ └── llm_client.py # LLM 客户端\n├── tests/\n└── requirements.txt", "section_ref": "29.2.1", "runnable": false, "dependencies": [] }, { "id": "code-3", "language": "python", "description": "", "code": "# app/config.py\n\"\"\"内容生成流水线配置\"\"\"\n\nfrom pydantic_settings import BaseSettings\nfrom enum import Enum\n\n\nclass WritingStyle(str, Enum):\n PROFESSIONAL = \"professional\" # 专业严谨\n CASUAL = \"casual\" # 轻松活泼\n STORYTELLING = \"storytelling\" # 故事叙述\n TUTORIAL = \"tutorial\" # 教程指南\n OPINION = \"opinion\" # 观点评论\n\n\nclass Channel(str, Enum):\n WECHAT = \"wechat\" # 微信公众号\n ZHIHU = \"zhihu\" # 知乎\n XIAOHONGSHU = \"xiaohongshu\" # 小红书\n TOUTIAO = \"toutiao\" # 头条号\n DOUYIN = \"douyin\" # 抖音脚本\n\n\nclass Settings(BaseSettings):\n APP_NAME: str = \"内容生成流水线\"\n APP_VERSION: str = \"1.0.0\"\n DEBUG: bool = False\n\n LLM_API_KEY: str = \"\"\n LLM_BASE_URL: str = \"https://api.openai.com/v1\"\n LLM_MODEL: str = \"gpt-4o\"\n LLM_MAX_TOKENS: int = 4096\n\n # 内容参数\n MAX_REVIEW_ITERATIONS: int = 3\n MIN_ORIGINALITY_SCORE: float = 0.85\n SENSITIVE_WORD_STRICT: bool = True\n\n # 渠道字数限制\n CHANNEL_LIMITS: dict = {\n \"wechat\": {\"min\": 1500, \"max\": 3000},\n \"zhihu\": {\"min\": 1000, \"max\": 5000},\n \"xiaohongshu\": {\"min\": 300, \"max\": 1000},\n \"toutiao\": {\"min\": 800, \"max\": 2500},\n \"douyin\": {\"min\": 100, \"max\": 500},\n }\n\n class Config:\n env_file = \".env\"\n env_prefix = \"CP_\"\n\n\nsettings = Settings()", "section_ref": "29.3.1", "runnable": true, "dependencies": [ "pydantic_settings" ] }, { "id": "code-4", "language": "python", "description": "settings = Settings()", "code": "# app/utils/llm_client.py\n\"\"\"LLM 客户端封装（同前章，略作简化）\"\"\"\n\nimport json\nfrom typing import Optional, List, Dict\nfrom openai import OpenAI\nfrom app.config import settings\n\n\nclass LLMClient:\n _instance: Optional['LLMClient'] = None\n\n def __new__(cls):\n if cls._instance is None:\n cls._instance = super().__new__(cls)\n cls._instance._client = OpenAI(\n api_key=settings.LLM_API_KEY,\n base_url=settings.LLM_BASE_URL)\n return cls._instance\n\n async def chat(self, messages, system_prompt=None,\n temperature=0.7, max_tokens=None) -> str:\n full = []\n if system_prompt:\n full.append({\"role\": \"system\", \"content\": system_prompt})\n full.extend(messages)\n resp = self._client.chat.completions.create(\n model=settings.LLM_MODEL, messages=full,\n temperature=temperature,\n max_tokens=max_tokens or settings.LLM_MAX_TOKENS)\n return resp.choices[0].message.content\n\n async def chat_json(self, messages, system_prompt=None) -> dict:\n content = await self.chat(messages, system_prompt,\n temperature=0.1,\n response_format={\"type\": \"json_object\"})\n return json.loads(content)\n\n\nllm_client = LLMClient()", "section_ref": "29.3.1", "runnable": true, "dependencies": [ "openai", "app" ] }, { "id": "code-5", "language": "python", "description": "", "code": "# app/agents/topic_agent.py\n\"\"\"选题策划 Agent\"\"\"\n\nimport json\nfrom dataclasses import dataclass, field\nfrom typing import List, Dict, Optional\nfrom datetime import datetime\nfrom app.utils.llm_client import llm_client\n\n\n@dataclass\nclass TopicSuggestion:\n title: str\n description: str\n target_audience: str\n keywords: List[str]\n estimated_read_time: int # 分钟\n trending_score: float # 热度评分 0-100\n competition_level: str # low/medium/high\n suggested_channels: List[str]\n outline: List[str] # 建议大纲\n\n\n@dataclass\nclass TopicAnalysis:\n suggestions: List[TopicSuggestion]\n trending_keywords: List[str]\n content_gaps: List[str]\n summary: str\n\n\nclass TopicAgent:\n SYSTEM_PROMPT = \"\"\"你是一个资深内容策划专家，擅长选题分析和内容规划。\n\n你的任务：\n1. 分析给定的领域和关键词，推荐 5 个高价值选题\n2. 为每个选题生成详细的内容大纲\n3. 评估选题的热度、竞争度和目标受众\n4. 推荐适合的分发渠道\n\n返回 JSON：\n{{\n \"suggestions\": [\n {{\n \"title\": \"文章标题\",\n \"description\": \"一句话描述\",\n \"target_audience\": \"目标受众\",\n \"keywords\": [\"关键词1\", \"关键词2\"],\n \"estimated_read_time\": 8,\n \"trending_score\": 85,\n \"competition_level\": \"medium\",\n \"suggested_channels\": [\"wechat\", \"zhihu\"],\n \"outline\": [\"一、引言：...\", \"二、核心观点：...\", ...]\n }}\n ],\n \"trending_keywords\": [\"趋势关键词\"],\n \"content_gaps\": [\"竞品尚未覆盖的内容方向\"],\n \"summary\": \"选题分析总结\"\n}}\"\"\"\n\n async def analyze(\n self,\n domain: str,\n keywords: Optional[List[str]] = None,\n target_channels: Optional[List[str]] = None,\n count: int = 5,\n ) -> TopicAnalysis:\n \"\"\"分析领域并推荐选题\"\"\"\n kw_text = f\"，关键词：{', '.join(keywords)}\" if keywords else \"\"\n ch_text = (f\"，目标渠道：{', '.join(target_channels)}\"\n if target_channels else \"\")\n\n messages = [{\n \"role\": \"user\",\n \"content\": (f\"请为以下领域推荐 {count} 个选题：\\n\"\n f\"领域：{domain}{kw_text}{ch_text}\\n\"\n f\"当前日期：{datetime.now().strftime('%Y-%m-%d')}\")\n }]\n\n try:\n result = await llm_client.chat_json(\n messages=messages, system_prompt=self.SYSTEM_PROMPT)\n suggestions = []\n for s in result.get(\"suggestions\", []):\n suggestions.append(TopicSuggestion(\n title=s.get(\"title\", \"\"),\n description=s.get(\"description\", \"\"),\n target_audience=s.get(\"target_audience\", \"\"),\n keywords=s.get(\"keywords\", []),\n estimated_read_time=s.get(\"estimated_read_time\", 5),\n trending_score=float(s.get(\"trending_score\", 50)),\n competition_level=s.get(\"competition_level\", \"medium\"),\n suggested_channels=s.get(\"suggested_channels\", []),\n outline=s.get(\"outline\", []),\n ))\n return TopicAnalysis(\n suggestions=suggestions,\n trending_keywords=result.get(\"trending_keywords\", []),\n content_gaps=result.get(\"content_gaps\", []),\n summary=result.get(\"summary\", \"\"),\n )\n except Exception as e:\n return TopicAnalysis(\n suggestions=[], trending_keywords=[],\n content_gaps=[], summary=f\"分析失败: {str(e)}\")\n\n async def refine_outline(\n self, topic: TopicSuggestion, feedback: str,\n ) -> TopicSuggestion:\n \"\"\"根据反馈优化大纲\"\"\"\n messages = [\n {\"role\": \"user\", \"content\": f\"选题: {topic.title}\\n\"\n f\"当前大纲: {topic.outline}\"},\n {\"role\": \"user\", \"content\": f\"优化反馈: {feedback}\\n\"\n \"请优化大纲结构。\"},\n ]\n try:\n result = await llm_client.chat_json(\n messages=messages, system_prompt=self.SYSTEM_PROMPT)\n if result.get(\"suggestions\"):\n s = result[\"suggestions\"][0]\n topic.outline = s.get(\"outline\", topic.outline)\n topic.description = s.get(\"description\", topic.description)\n except Exception:\n pass\n return topic", "section_ref": "29.3.2", "runnable": true, "dependencies": [ "app" ] }, { "id": "code-6", "language": "python", "description": "", "code": "# app/agents/writing_agent.py\n\"\"\"内容撰写 Agent\"\"\"\n\nfrom dataclasses import dataclass, field\nfrom typing import List, Dict, Optional\nfrom app.utils.llm_client import llm_client\nfrom app.agents.topic_agent import TopicSuggestion\nfrom app.config import WritingStyle\n\n\n@dataclass\nclass ArticleDraft:\n title: str\n subtitle: str = \"\"\n content: str = \"\"\n sections: List[Dict[str, str]] = field(default_factory=list)\n # sections: [{\"heading\": \"...\", \"body\": \"...\"}]\n word_count: int = 0\n style: str = \"professional\"\n target_channel: str = \"wechat\"\n references: List[str] = field(default_factory=list)\n\n\nSTYLE_GUIDES = {\n WritingStyle.PROFESSIONAL: {\n \"tone\": \"专业、严谨、有深度\",\n \"structure\": \"提出问题→分析原因→给出方案→总结展望\",\n \"avoid\": [\"口语化表达\", \"过度感叹号\", \"网络流行语\"],\n },\n WritingStyle.CASUAL: {\n \"tone\": \"轻松、有趣、接地气\",\n \"structure\": \"吸引眼球开头→轻松展开→互动结尾\",\n \"avoid\": [\"过于学术的术语\", \"长篇大论\", \"干巴巴的数据\"],\n },\n WritingStyle.STORYTELLING: {\n \"tone\": \"故事化、场景化、有画面感\",\n \"structure\": \"场景引入→人物/事件→冲突→解决→升华\",\n \"avoid\": [\"空洞说教\", \"枯燥理论\", \"没有场景\"],\n },\n WritingStyle.TUTORIAL: {\n \"tone\": \"清晰、步骤化、实用\",\n \"structure\": \"需求场景→环境准备→步骤详解→常见问题\",\n \"avoid\": [\"模糊描述\", \"跳过关键步骤\", \"缺少代码示例\"],\n },\n WritingStyle.OPINION: {\n \"tone\": \"犀利、有观点、引发思考\",\n \"structure\": \"热点切入→核心观点→论证→对立观点→结论\",\n \"avoid\": [\"模棱两可\", \"人云亦云\", \"没有数据支撑\"],\n },\n}\n\n\nclass WritingAgent:\n def __init__(self):\n pass\n\n async def write(\n self,\n topic: TopicSuggestion,\n style: WritingStyle = WritingStyle.PROFESSIONAL,\n target_channel: str = \"wechat\",\n reference_materials: Optional[List[str]] = None,\n max_words: int = 2500,\n ) -> ArticleDraft:\n \"\"\"根据选题和大纲撰写内容\"\"\"\n style_guide = STYLE_GUIDES[style]\n\n system_prompt = f\"\"\"你是一个专业的内容创作者。\n\n写作风格要求：\n- 语气：{style_guide['tone']}\n- 结构：{style_guide['structure']}\n- 避免：{', '.join(style_guide['avoid'])}\n- 目标平台：{target_channel}\n\n写作规范：\n1. 每个段落控制在 100-200 字\n2. 使用小标题分割内容模块\n3. 适当使用数据、案例、引用增强说服力\n4. 结尾要有行动号召或互动引导\n5. 目标字数：{max_words}字左右\n\n返回 JSON：\n{{\n \"title\": \"最终标题\",\n \"subtitle\": \"副标题\",\n \"sections\": [\n {{\"heading\": \"小标题\", \"body\": \"正文内容\"}},\n ...\n ],\n \"references\": [\"引用的资料来源\"]\n}}\"\"\"\n\n # 构建参考资料\n ref_text = \"\"\n if reference_materials:\n ref_text = \"\\n\\n参考资料：\\n\" + \"\\n---\\n\".join(reference_materials)\n\n messages = [{\n \"role\": \"user\",\n \"content\": (f\"请撰写以下文章：\\n\\n\"\n f\"选题：{topic.title}\\n\"\n f\"描述：{topic.description}\\n\"\n f\"目标受众：{topic.target_audience}\\n\"\n f\"关键词：{', '.join(topic.keywords)}\\n\"\n f\"大纲：\\n\" +\n \"\\n\".join(f\" {i+1}. {item}\"\n for i, item in enumerate(topic.outline)) +\n ref_text)\n }]\n\n try:\n import json\n result = await llm_client.chat_json(\n messages=messages, system_prompt=system_prompt,\n temperature=0.7, max_tokens=4096)\n\n sections = []\n for s in result.get(\"sections\", []):\n sections.append({\n \"heading\": s.get(\"heading\", \"\"),\n \"body\": s.get(\"body\", \"\"),\n })\n\n full_content = \"\\n\\n\".join(\n f\"## {s['heading']}\\n\\n{s['body']}\" if s['heading']\n else s['body']\n for s in sections\n )\n word_count = len(full_content)\n\n return ArticleDraft(\n title=result.get(\"title\", topic.title),\n subtitle=result.get(\"subtitle\", \"\"),\n content=full_content,\n sections=sections,\n word_count=word_count,\n style=style.value,\n target_channel=target_channel,\n references=result.get(\"references\", []),\n )\n except Exception as e:\n return ArticleDraft(\n title=topic.title, content=f\"撰写失败: {str(e)}\")\n\n async def rewrite_section(\n self, section: Dict[str, str], feedback: str,\n style: WritingStyle = WritingStyle.PROFESSIONAL,\n ) -> Dict[str, str]:\n \"\"\"根据审校反馈重写某个段落\"\"\"\n style_guide = STYLE_GUIDES[style]\n system_prompt = f\"\"\"你是一个内容编辑，根据反馈优化段落。\n风格要求：{style_guide['tone']}\n\n返回 JSON：{{\"heading\": \"...\", \"body\": \"...\"}}\"\"\"\n\n messages = [\n {\"role\": \"user\", \"content\": f\"原段落标题: {section['heading']}\\n\"\n f\"原段落内容: {section['body']}\"},\n {\"role\": \"user\", \"content\": f\"修改意见: {feedback}\"},\n ]\n\n try:\n import json\n result = await llm_client.chat_json(\n messages=messages, system_prompt=system_prompt,\n temperature=0.7)\n return {\n \"heading\": result.get(\"heading\", section[\"heading\"]),\n \"body\": result.get(\"body\", section[\"body\"]),\n }\n except Exception:\n return section", "section_ref": "29.3.3", "runnable": true, "dependencies": [ "app" ] }, { "id": "code-7", "language": "python", "description": "", "code": "# app/agents/review_agent.py\n\"\"\"审校优化 Agent\"\"\"\n\nimport re\nfrom dataclasses import dataclass, field\nfrom typing import List, Dict, Optional, Tuple\nfrom app.utils.llm_client import llm_client\nfrom app.agents.writing_agent import ArticleDraft\n\n\n@dataclass\nclass ReviewIssue:\n section_index: int\n issue_type: str # grammar/style/seo/sensitive/readability\n severity: str # info/warning/error\n description: str\n suggestion: str\n auto_fixable: bool = False\n\n\n@dataclass\nclass ReviewResult:\n overall_score: float # 0-100\n issues: List[ReviewIssue]\n readability_score: float\n seo_score: float\n sensitivity_check: Dict[str, List[str]] # {\"pass\": [], \"fail\": []}\n word_count: int\n keyword_density: Dict[str, float]\n summary: str\n needs_rewrite: bool = False\n\n\nclass ReviewAgent:\n \"\"\"多维度内容审校\"\"\"\n\n # 基础敏感词库（生产环境应使用专业词库）\n SENSITIVE_WORDS = [\n \"最\", \"第一\", \"首个\", \"独家\", \"国家级\", \"世界级\",\n \"100%\", \"绝对\", \"保证\", \"包治\", \"根治\",\n ]\n\n def review(\n self,\n draft: ArticleDraft,\n keywords: Optional[List[str]] = None,\n ) -> ReviewResult:\n \"\"\"执行多维度审校\"\"\"\n issues = []\n\n # 1. 敏感词检查\n sensitivity = self._check_sensitive(draft.content)\n for word in sensitivity[\"fail\"]:\n issues.append(ReviewIssue(\n section_index=-1, issue_type=\"sensitive\",\n severity=\"error\",\n description=f\"发现敏感词: {word}\",\n suggestion=f\"替换或删除「{word}」\",\n auto_fixable=False,\n ))\n\n # 2. 可读性评分\n readability = self._score_readability(draft.content)\n\n # 3. SEO 评分\n seo_score, keyword_density = self._score_seo(\n draft.content, draft.title, keywords or [])\n\n # 4. LLM 深度审校（语法、风格、逻辑）\n llm_issues = self._llm_review(draft)\n issues.extend(llm_issues)\n\n # 5. 字数检查\n channel_limits = settings.CHANNEL_LIMITS.get(\n draft.target_channel, {\"min\": 500, \"max\": 5000})\n if draft.word_count < channel_limits[\"min\"]:\n issues.append(ReviewIssue(\n section_index=-1, issue_type=\"readability\",\n severity=\"warning\",\n description=f\"字数不足: {draft.word_count}字，\"\n f\"建议最少{channel_limits['min']}字\",\n suggestion=\"增加更多论据和案例\",\n ))\n if draft.word_count > channel_limits[\"max\"]:\n issues.append(ReviewIssue(\n section_index=-1, issue_type=\"readability\",\n severity=\"warning\",\n description=f\"字数超标: {draft.word_count}字，\"\n f\"建议最多{channel_limits['max']}字\",\n suggestion=\"精简内容，删减冗余段落\",\n ))\n\n # 计算总分\n overall = self._calculate_overall(\n readability, seo_score,\n len(sensitivity[\"fail\"]) == 0,\n len([i for i in issues if i.severity == \"error\"]) == 0)\n\n needs_rewrite = any(i.severity == \"error\" for i in issues)\n\n return ReviewResult(\n overall_score=overall,\n issues=issues,\n readability_score=readability,\n seo_score=seo_score,\n sensitivity_check=sensitivity,\n word_count=draft.word_count,\n keyword_density=keyword_density,\n summary=self._generate_summary(issues, overall),\n needs_rewrite=needs_rewrite,\n )\n\n def _check_sensitive(self, text: str) -> Dict[str, List[str]]:\n \"\"\"敏感词检查\"\"\"\n found_pass = []\n found_fail = []\n for word in self.SENSITIVE_WORDS:\n if word in text:\n # 广告法极限词需要严格处理\n if word in [\"最\", \"第一\", \"首个\", \"100%\", \"绝对\", \"国家级\"]:\n found_fail.append(word)\n else:\n found_pass.append(word)\n return {\"pass\": found_pass, \"fail\": found_fail}\n\n def _score_readability(self, text: str) -> float:\n \"\"\"可读性评分（0-100）\"\"\"\n score = 70.0 # 基础分\n\n # 段落长度适中（100-300字最佳）\n paragraphs = [p.strip() for p in text.split(\"\\n\\n\") if p.strip()]\n for p in paragraphs:\n length = len(p)\n if 100 <= length <= 300:\n score += 2\n elif length > 500:\n score -= 5\n\n # 句子长度适中（10-30字最佳）\n sentences = re.split(r'[。！？；]', text)\n avg_len = sum(len(s) for s in sentences) / max(len(sentences), 1)\n if 10 <= avg_len <= 30:\n score += 10\n elif avg_len > 50:\n score -= 10\n\n # 小标题分布\n headings = re.findall(r'^##?\\s+.+$', text, re.MULTILINE)\n if len(headings) >= 3:\n score += 5\n\n # 互动元素（提问、感叹）\n if re.search(r'[？?!]', text):\n score += 3\n\n return min(max(score, 0), 100)\n\n def _score_seo(\n self, text: str, title: str, keywords: List[str],\n ) -> Tuple[float, Dict[str, float]]:\n \"\"\"SEO 评分\"\"\"\n score = 60.0\n density = {}\n\n if not keywords:\n return score, density\n\n total_words = len(text)\n for kw in keywords:\n count = text.count(kw)\n d = count / max(total_words, 1) * 100\n density[kw] = round(d, 2)\n # 理想关键词密度 1%-3%\n if 1.0 <= d <= 3.0:\n score += 8\n elif d > 5.0:\n score -= 5 # 关键词堆砌扣分\n # 标题包含关键词\n if kw in title:\n score += 5\n\n return min(max(score, 0), 100), density\n\n def _llm_review(self, draft: ArticleDraft) -> List[ReviewIssue]:\n \"\"\"LLM 深度审校\"\"\"\n # 逐段审校（限制段落数避免 Token 过多）\n issues = []\n for i, section in enumerate(draft.sections[:5]):\n try:\n import json\n result = self._review_section(section)\n issues.append(ReviewIssue(\n section_index=i,\n issue_type=result.get(\"type\", \"style\"),\n severity=result.get(\"severity\", \"info\"),\n description=result.get(\"description\", \"\"),\n suggestion=result.get(\"suggestion\", \"\"),\n auto_fixable=result.get(\"auto_fixable\", False),\n ))\n except Exception:\n pass\n return issues\n\n def _review_section(self, section: Dict) -> Dict:\n \"\"\"审校单个段落\"\"\"\n import json\n import asyncio\n system_prompt = \"\"\"审校以下段落，检查：\n1. 语法错误\n2. 逻辑不通顺\n3. 表达不清晰\n4. 事实性错误风险\n\n返回 JSON：\n{\"type\": \"grammar|style|logic\", \"severity\": \"info|warning|error\",\n \"description\": \"问题描述\", \"suggestion\": \"修改建议\",\n \"auto_fixable\": false}\n如果没有问题返回：\n{\"type\": \"pass\", \"severity\": \"info\",\n \"description\": \"无问题\", \"suggestion\": \"\", \"auto_fixable\": false}\"\"\"\n\n # 简化处理，实际可用 async\n result = llm_client._client.chat.completions.create(\n model=settings.LLM_MODEL,\n messages=[\n {\"role\": \"system\", \"content\": system_prompt},\n {\"role\": \"user\",\n \"content\": f\"标题: {section['heading']}\\n\"\n f\"内容: {section['body'][:500]}\"},\n ],\n temperature=0.1,\n max_tokens=500,\n )\n return json.loads(result.choices[0].message.content)\n\n def _calculate_overall(\n self, readability, seo, sensitive_pass, no_errors,\n ) -> float:\n \"\"\"计算综合评分\"\"\"\n score = readability * 0.3 + seo * 0.3 + 40\n if not sensitive_pass:\n score -= 20\n if not no_errors:\n score -= 10\n return min(max(round(score, 1), 0), 100)\n\n def _generate_summary(\n self, issues: List[ReviewIssue], score: float,\n ) -> str:\n errors = [i for i in issues if i.severity == \"error\"]\n warnings = [i for i in issues if i.severity == \"warning\"]\n parts = [f\"综合评分: {score}/100\"]\n if errors:\n parts.append(f\"❌ {len(errors)} 个严重问题需要修改\")\n if warnings:\n parts.append(f\"⚠️ {len(warnings)} 个建议优化项\")\n if not errors and not warnings:\n parts.append(\"✅ 审校通过，质量良好\")\n return \"；\".join(parts)", "section_ref": "29.3.4", "runnable": true, "dependencies": [ "app" ] }, { "id": "code-8", "language": "python", "description": "", "code": "# app/agents/distribute_agent.py\n\"\"\"多渠道分发适配 Agent\"\"\"\n\nfrom dataclasses import dataclass, field\nfrom typing import List, Dict, Optional\nfrom app.utils.llm_client import llm_client\nfrom app.agents.writing_agent import ArticleDraft\nfrom app.config import Channel\n\n\n@dataclass\nclass ChannelContent:\n channel: str\n title: str\n content: str\n tags: List[str]\n word_count: int\n adaptation_notes: str\n\n\nCHANNEL_STYLES = {\n Channel.WECHAT: {\n \"name\": \"微信公众号\",\n \"format\": \"Markdown → 排版\",\n \"title_style\": \"吸引眼球但不过标题党，20字以内\",\n \"content_style\": \"分段清晰、图文并茂、引用用引用框\",\n \"tone\": \"专业但有温度\",\n \"max_length\": 3000,\n \"features\": [\"文末引导关注\", \"阅读原文链接\", \"分享提示\"],\n },\n Channel.ZHIHU: {\n \"name\": \"知乎\",\n \"format\": \"Markdown\",\n \"title_style\": \"提问式或观点式，体现专业性\",\n \"content_style\": \"逻辑严密、数据支撑、引用来源\",\n \"tone\": \"理性、深度、有干货\",\n \"max_length\": 5000,\n \"features\": [\"开头直击痛点\", \"结构化论证\", \"评论区互动\"],\n },\n Channel.XIAOHONGSHU: {\n \"name\": \"小红书\",\n \"format\": \"短文 + Emoji\",\n \"title_style\": \"20字以内，含关键词和Emoji\",\n \"content_style\": \"要点式、多用Emoji、话题标签\",\n \"tone\": \"种草感、闺蜜式分享\",\n \"max_length\": 1000,\n \"features\": [\"吸睛首图\", \"话题标签 #\", \"互动引导\"],\n },\n Channel.TOUTIAO: {\n \"name\": \"头条号\",\n \"format\": \"富文本\",\n \"title_style\": \"信息量大、含数字和关键词\",\n \"content_style\": \"新闻式开头、结构化内容\",\n \"tone\": \"客观、信息量大\",\n \"max_length\": 2500,\n \"features\": [\"三段式标题\", \"摘要精炼\", \"标签分类\"],\n },\n Channel.DOUYIN: {\n \"name\": \"抖音脚本\",\n \"format\": \"视频脚本\",\n \"title_style\": \"口语化、有悬念\",\n \"content_style\": \"口语化表达、场景描述、时间标注\",\n \"tone\": \"口语化、快节奏、有画面感\",\n \"max_length\": 500,\n \"features\": [\"前3秒吸引\", \"节奏标注\", \"BGM建议\"],\n },\n}\n\n\nclass DistributeAgent:\n SYSTEM_PROMPT = \"\"\"你是一个内容分发专家，擅长将一篇文章适配为不同平台的风格。\n\n规则：\n1. 保留核心观点和信息\n2. 调整标题、结构和语气以匹配目标平台\n3. 添加平台特有的元素（话题标签、引导语等）\n4. 控制字数在目标范围内\n\n返回 JSON：\n{{\n \"title\": \"适配后的标题\",\n \"content\": \"适配后的正文\",\n \"tags\": [\"标签1\", \"标签2\"],\n \"adaptation_notes\": \"适配说明\"\n}}\"\"\"\n\n async def adapt(\n self, draft: ArticleDraft, channels: List[str],\n ) -> List[ChannelContent]:\n \"\"\"并行适配多渠道内容\"\"\"\n import asyncio\n tasks = [\n self._adapt_single(draft, ch) for ch in channels\n ]\n results = await asyncio.gather(*tasks, return_exceptions=True)\n return [r for r in results if isinstance(r, ChannelContent)]\n\n async def _adapt_single(\n self, draft: ArticleDraft, channel: str,\n ) -> ChannelContent:\n \"\"\"适配单个渠道\"\"\"\n style = CHANNEL_STYLES.get(Channel(channel), CHANNEL_STYLES[Channel.WECHAT])\n\n messages = [{\n \"role\": \"user\",\n \"content\": (f\"请将以下文章适配为【{style['name']}】平台：\\n\\n\"\n f\"原标题: {draft.title}\\n\"\n f\"原内容: {draft.content[:2000]}\\n\\n\"\n f\"要求:\\n\"\n f\"- 标题风格: {style['title_style']}\\n\"\n f\"- 内容风格: {style['content_style']}\\n\"\n f\"- 语气: {style['tone']}\\n\"\n f\"- 最大字数: {style['max_length']}\\n\"\n f\"- 特色: {', '.join(style['features'])}\")\n }]\n\n try:\n import json\n result = await llm_client.chat_json(\n messages=messages, system_prompt=self.SYSTEM_PROMPT,\n temperature=0.6)\n content = result.get(\"content\", \"\")\n return ChannelContent(\n channel=channel,\n title=result.get(\"title\", draft.title),\n content=content,\n tags=result.get(\"tags\", []),\n word_count=len(content),\n adaptation_notes=result.get(\"adaptation_notes\", \"\"),\n )\n except Exception as e:\n return ChannelContent(\n channel=channel, title=draft.title,\n content=draft.content, tags=[],\n word_count=draft.word_count,\n adaptation_notes=f\"适配失败: {str(e)}\")", "section_ref": "29.3.5", "runnable": true, "dependencies": [ "app" ] }, { "id": "code-9", "language": "python", "description": "", "code": "# app/services/sensitive_filter.py\n\"\"\"敏感词过滤服务\"\"\"\n\nimport re\nfrom dataclasses import dataclass, field\nfrom typing import List, Dict, Tuple\n\n\n# 广告法极限词（基础版）\nAD_LIMIT_WORDS = [\n \"最\", \"最佳\", \"最好\", \"最优\", \"最强\", \"最先进\",\n \"第一\", \"首个\", \"首款\", \"唯一\", \"独一无二\",\n \"国家级\", \"世界级\", \"顶级\", \"极致\", \"绝无仅有\",\n \"100%\", \"绝对\", \"保证治愈\", \"包治百病\", \"根治\",\n \"永久\", \"万能\", \"特效\", \"高效\", \"超效\",\n]\n\n\n# 政治敏感词（基础版，生产环境需要完整词库）\nPOLITICAL_WORDS = [\n # 生产环境请使用专业敏感词库\n]\n\n\n@dataclass\nclass FilterResult:\n is_clean: bool\n found_words: List[Tuple[str, int]] # (word, position)\n suggestions: List[str]\n cleaned_text: str = \"\"\n\n\nclass SensitiveFilter:\n def __init__(self, extra_words: List[str] = None):\n self._words = set(AD_LIMIT_WORDS + POLITICAL_WORDS)\n if extra_words:\n self._words.update(extra_words)\n # 按长度降序排列，优先匹配长词\n self._sorted_words = sorted(self._words, key=len, reverse=True)\n\n def check(self, text: str) -> FilterResult:\n \"\"\"检查文本中的敏感词\"\"\"\n found = []\n for word in self._sorted_words:\n idx = 0\n while True:\n pos = text.find(word, idx)\n if pos == -1:\n break\n found.append((word, pos))\n idx = pos + len(word)\n\n # 生成替换建议\n suggestions = []\n replacements = {\n \"最\": \"非常\", \"最佳\": \"优秀\", \"第一\": \"领先\",\n \"100%\": \"极高\", \"绝对\": \"非常\",\n }\n for word, pos in found:\n replacement = replacements.get(word, f\"**{word[0]}{'*'*(len(word)-1)}\")\n suggestions.append(f\"位置{pos}: 「{word}」→ 建议替换为「{replacement}」\")\n\n # 自动替换\n cleaned = text\n for word, _ in found:\n replacement = replacements.get(word, \"\")\n if replacement:\n cleaned = cleaned.replace(word, replacement)\n\n return FilterResult(\n is_clean=len(found) == 0,\n found_words=found,\n suggestions=suggestions,\n cleaned_text=cleaned if found else text,\n )\n\n def check_title(self, title: str) -> FilterResult:\n \"\"\"标题敏感词检查（更严格）\"\"\"\n return self.check(title)", "section_ref": "29.3.6", "runnable": true, "dependencies": [] }, { "id": "code-10", "language": "python", "description": "", "code": "# app/main.py\n\"\"\"内容生成流水线 - FastAPI 入口\"\"\"\n\nfrom contextlib import asynccontextmanager\nfrom fastapi import FastAPI, HTTPException\nfrom fastapi.middleware.cors import CORSMiddleware\nfrom pydantic import BaseModel\nfrom typing import Optional, List\n\nfrom app.config import settings, WritingStyle\nfrom app.agents.topic_agent import TopicAgent\nfrom app.agents.writing_agent import WritingAgent\nfrom app.agents.review_agent import ReviewAgent\nfrom app.agents.distribute_agent import DistributeAgent\nfrom app.services.sensitive_filter import SensitiveFilter\n\ntopic_agent = TopicAgent()\nwriting_agent = WritingAgent()\nreview_agent = ReviewAgent()\ndistribute_agent = DistributeAgent()\nsensitive_filter = SensitiveFilter()\n\n\n@asynccontextmanager\nasync def lifespan(app: FastAPI):\n print(f\"🚀 {settings.APP_NAME} v{settings.APP_VERSION} 启动\")\n yield\n\n\napp = FastAPI(\n title=settings.APP_NAME, version=settings.APP_VERSION,\n lifespan=lifespan,\n description=\"AI 驱动的内容生成流水线\")\napp.add_middleware(CORSMiddleware, allow_origins=[\"*\"],\n allow_credentials=True,\n allow_methods=[\"*\"], allow_headers=[\"*\"])\n\n\nclass TopicRequest(BaseModel):\n domain: str\n keywords: Optional[List[str]] = None\n channels: Optional[List[str]] = None\n count: int = 5\n\n\nclass WriteRequest(BaseModel):\n topic_title: str\n topic_description: str\n topic_keywords: List[str]\n outline: List[str]\n target_audience: str = \"通用\"\n style: str = \"professional\"\n channel: str = \"wechat\"\n max_words: int = 2500\n\n\nclass PipelineRequest(BaseModel):\n domain: str\n keywords: Optional[List[str]] = None\n style: str = \"professional\"\n channels: List[str] = [\"wechat\", \"zhihu\"]\n max_words: int = 2500\n\n\n@app.get(\"/health\")\nasync def health():\n return {\"status\": \"ok\", \"version\": settings.APP_VERSION}\n\n\n@app.post(\"/api/v1/topics\")\nasync def analyze_topics(req: TopicRequest):\n \"\"\"分析领域并推荐选题\"\"\"\n result = await topic_agent.analyze(\n domain=req.domain,\n keywords=req.keywords,\n target_channels=req.channels,\n count=req.count,\n )\n return {\n \"suggestions\": [\n {\"title\": s.title, \"description\": s.description,\n \"trending_score\": s.trending_score,\n \"competition_level\": s.competition_level,\n \"keywords\": s.keywords,\n \"suggested_channels\": s.suggested_channels,\n \"outline\": s.outline}\n for s in result.suggestions\n ],\n \"trending_keywords\": result.trending_keywords,\n \"content_gaps\": result.content_gaps,\n \"summary\": result.summary,\n }\n\n\n@app.post(\"/api/v1/write\")\nasync def write_article(req: WriteRequest):\n \"\"\"根据选题撰写文章\"\"\"\n from app.agents.topic_agent import TopicSuggestion\n topic = TopicSuggestion(\n title=req.topic_title,\n description=req.topic_description,\n target_audience=req.target_audience,\n keywords=req.topic_keywords,\n outline=req.outline,\n )\n style = WritingStyle(req.style)\n draft = await writing_agent.write(\n topic=topic, style=style,\n target_channel=req.channel,\n max_words=req.max_words,\n )\n return {\n \"title\": draft.title,\n \"subtitle\": draft.subtitle,\n \"content\": draft.content,\n \"sections\": draft.sections,\n \"word_count\": draft.word_count,\n \"style\": draft.style,\n }\n\n\n@app.post(\"/api/v1/review\")\nasync def review_article(\n content: str,\n title: str = \"\",\n keywords: Optional[List[str]] = None,\n):\n \"\"\"审校文章\"\"\"\n draft = ArticleDraft(title=title, content=content,\n word_count=len(content))\n result = review_agent.review(draft, keywords)\n return {\n \"overall_score\": result.overall_score,\n \"readability_score\": result.readability_score,\n \"seo_score\": result.seo_score,\n \"needs_rewrite\": result.needs_rewrite,\n \"issues\": [\n {\"section\": i.issue_type,\n \"severity\": i.severity,\n \"description\": i.description,\n \"suggestion\": i.suggestion}\n for i in result.issues\n ],\n \"sensitive_check\": result.sensitivity_check,\n \"summary\": result.summary,\n }\n\n\n@app.post(\"/api/v1/pipeline\")\nasync def run_pipeline(req: PipelineRequest):\n \"\"\"完整流水线：选题 → 撰写 → 审校 → 分发\"\"\"\n # 1. 选题分析\n topic_result = await topic_agent.analyze(\n domain=req.domain,\n keywords=req.keywords,\n target_channels=req.channels,\n count=1,\n )\n if not topic_result.suggestions:\n raise HTTPException(400, detail=\"未能生成合适选题\")\n\n topic = topic_result.suggestions[0]\n\n # 2. 撰写初稿\n style = WritingStyle(req.style)\n draft = await writing_agent.write(\n topic=topic, style=style,\n target_channel=req.channels[0],\n max_words=req.max_words,\n )\n\n # 3. 审校\n review = review_agent.review(draft, topic.keywords)\n\n # 4. 迭代修改（如有严重问题）\n if review.needs_rewrite and settings.MAX_REVIEW_ITERATIONS > 0:\n errors = [i for i in review.issues if i.severity == \"error\"]\n for issue in errors[:2]: # 最多修2个问题\n if 0 <= issue.section_index < len(draft.sections):\n draft.sections[issue.section_index] = \\\n await writing_agent.rewrite_section(\n draft.sections[issue.section_index],\n issue.suggestion, style)\n\n # 5. 多渠道适配\n adapted = await distribute_agent.adapt(draft, req.channels)\n\n return {\n \"topic\": {\n \"title\": topic.title,\n \"description\": topic.description,\n \"trending_score\": topic.trending_score,\n },\n \"draft\": {\n \"title\": draft.title,\n \"content\": draft.content[:500] + \"...\",\n \"word_count\": draft.word_count,\n },\n \"review\": {\n \"overall_score\": review.overall_score,\n \"readability\": review.readability_score,\n \"seo\": review.seo_score,\n \"summary\": review.summary,\n },\n \"distributed\": [\n {\"channel\": a.channel, \"title\": a.title,\n \"word_count\": a.word_count, \"tags\": a.tags}\n for a in adapted\n ],\n }\n\n\nif __name__ == \"__main__\":\n import uvicorn\n uvicorn.run(\"app.main:app\", host=\"0.0.0.0\", port=8000,\n reload=settings.DEBUG)", "section_ref": "29.3.7", "runnable": true, "dependencies": [ "contextlib", "fastapi", "pydantic", "app" ] }, { "id": "code-11", "language": "python", "description": "", "code": "# tests/test_topic.py\n\"\"\"选题分析测试\"\"\"\n\nimport pytest\nfrom app.agents.topic_agent import TopicAgent\n\n\n@pytest.mark.asyncio\nasync def test_topic_analysis():\n agent = TopicAgent()\n # 需要配置 LLM API_KEY\n # result = await agent.analyze(\"AI技术\", [\"大模型\", \"Agent\"])\n # assert len(result.suggestions) > 0\n # assert result.suggestions[0].title\n pass", "section_ref": "29.4.1", "runnable": true, "dependencies": [ "pytest", "app" ] }, { "id": "code-12", "language": "python", "description": "", "code": "# tests/test_sensitive.py\n\"\"\"敏感词过滤测试\"\"\"\n\nfrom app.services.sensitive_filter import SensitiveFilter\n\n\ndef test_ad_limit_words():\n sf = SensitiveFilter()\n text = \"这是全国最好用的产品，100%有效，绝对保证\"\n result = sf.check(text)\n assert not result.is_clean\n assert len(result.found_words) >= 3\n\n\ndef test_clean_text():\n sf = SensitiveFilter()\n text = \"这是一篇关于技术分享的文章\"\n result = sf.check(text)\n assert result.is_clean\n assert len(result.found_words) == 0\n\n\ndef test_title_check():\n sf = SensitiveFilter()\n title = \"全球第一款AI写作工具\"\n result = sf.check_title(title)\n assert not result.is_clean", "section_ref": "29.4.2", "runnable": true, "dependencies": [ "app" ] }, { "id": "code-13", "language": "python", "description": "", "code": "# tests/test_readability.py\n\"\"\"可读性评分测试\"\"\"\n\nfrom app.agents.review_agent import ReviewAgent\nfrom app.agents.writing_agent import ArticleDraft\n\n\ndef test_good_readability():\n agent = ReviewAgent()\n text = (\"## 引言\\n\\nAI技术的发展日新月异。\\n\\n\"\n \"你知道Agent是什么吗？它可能改变我们的工作方式。\\n\\n\"\n \"## 核心观点\\n\\n首先，Agent具备自主决策能力。\\n\"\n \"其次，Agent可以调用外部工具。\\n\"\n \"最后，Agent能够从反馈中学习。\")\n draft = ArticleDraft(title=\"test\", content=text, word_count=len(text))\n score = agent._score_readability(text)\n assert score > 60\n\n\ndef test_long_paragraph_penalty():\n agent = ReviewAgent()\n text = \"这是一段非常长的文字\" * 100\n score = agent._score_readability(text)\n assert score < 70", "section_ref": "29.4.3", "runnable": true, "dependencies": [ "app" ] }, { "id": "code-14", "language": "dockerfile", "description": "---", "code": "FROM python:3.11-slim\nWORKDIR /app\nCOPY requirements.txt .\nRUN pip install --no-cache-dir -r requirements.txt\nCOPY app/ ./app/\nEXPOSE 8000\nCMD [\"uvicorn\", \"app.main:app\", \"--host\", \"0.0.0.0\", \"--port\", \"8000\"]", "section_ref": "29.5", "runnable": false, "dependencies": [] }, { "id": "code-15", "language": "yaml", "description": "CMD [\"uvicorn\", \"app.main:app\", \"--host\", \"0.0.0.0\", \"--port\", \"8000\"]", "code": "# docker-compose.yml\nversion: '3.8'\nservices:\n content-pipeline:\n build: .\n ports: [\"8000:8000\"]\n environment:\n - CP_LLM_API_KEY=${LLM_API_KEY}", "section_ref": "29.5", "runnable": false, "dependencies": [] } ], "tables": [ { "headers": [ "维度", "指标" ], "data": [ [ "单篇文章生成时间", "< 30 秒（初稿）" ], [ "内容原创度", "> 85%（相似度检测）" ], [ "敏感词准确率", "> 95%" ], [ "并发支持", "50 QPS" ], [ "支持渠道", "微信公众号、知乎、小红书、头条号、抖音" ] ] }, { "headers": [ "模式", "应用场景", "效果" ], "data": [ [ "Style Persona", "内容风格控制", "AI味降低 80%" ], [ "规则 + LLM 混合审校", "内容质量控制", "成本降低 60%" ], [ "Pipeline + 反馈循环", "内容质量迭代", "一次通过率 70%→90%" ], [ "渠道模板隔离", "多渠道适配", "各平台质量达标率 95%" ] ] } ], "key_takeaways": [], "common_pitfalls": [], "related_chapters": [ "ch18" ] }