#!/usr/bin/env python3
"""
平台适配文章撰写
根据大纲和平台配置（字数/格式/配图要求），为知乎/公众号/小红书各平台生成适配内容
"""
import json, datetime, logging, sys, re
from pathlib import Path
from typing import Dict, List, Optional
PROJECT_ROOT = Path(__file__).parent.parent
sys.path.insert(0, str(PROJECT_ROOT))
sys.path.insert(0, str(PROJECT_ROOT / 'platform' / 'backend'))
from db_helper import get_topic_by_id, update_topic_status, save_article
from content_cleaner import strip_thinking, strip_ai_preface, clean_markdown_content, clean_html_content
from prompt_loader import get_prompt, get_prompt_params
from image_generator import insert_lead_image
try:
from app.core.nvidia_client import call_llm
HAVE_LLM = True
except ImportError:
HAVE_LLM = False
import mistune
DATA_DIR = PROJECT_ROOT / "automation" / "data"
OUTLINE_DIR = DATA_DIR / "outlines"
TEMPLATES_DIR = PROJECT_ROOT / "automation" / "templates"
LOGS_DIR = PROJECT_ROOT / "automation" / "logs"
TODAY = datetime.datetime.now().strftime("%Y-%m-%d")
GEN_TIME = datetime.datetime.now().strftime("%Y-%m-%d %H:%M:%S")
logging.basicConfig(
level=logging.INFO,
format='%(asctime)s - %(levelname)s - %(message)s',
handlers=[
logging.FileHandler(LOGS_DIR / f"writer_{TODAY}.log"),
logging.StreamHandler()
]
)
logger = logging.getLogger(__name__)
_md_parser = mistune.create_markdown()
_FALLBACK_PLATFORM_CONFIG = {
"zhihu": {"max_chars": 8000, "style": "深度长文分析", "min_chars": 3000},
"wechat": {"max_chars": 4000, "style": "个人叙事对话感", "min_chars": 2000},
"xiaohongshu": {"max_chars": 1000, "style": "图文笔记，精炼实用", "min_chars": 400},
}
def _load_platform_config() -> dict:
try:
from app.database import SessionLocal
from app.models import PlatformConfig
db = SessionLocal()
configs = db.query(PlatformConfig).all()
db.close()
result = {}
for c in configs:
result[c.platform] = {
"max_chars": c.max_words or _FALLBACK_PLATFORM_CONFIG.get(c.platform, {}).get("max_chars", 3000),
"style": c.default_format or _FALLBACK_PLATFORM_CONFIG.get(c.platform, {}).get("style", "深度内容"),
"min_chars": c.min_words or _FALLBACK_PLATFORM_CONFIG.get(c.platform, {}).get("min_chars", 300),
}
return result
except Exception:
pass
return dict(_FALLBACK_PLATFORM_CONFIG)
PLATFORM_CONFIG = _load_platform_config()
PLATFORM_NAMES = {
"zhihu": "知乎专栏",
"wechat": "微信公众号",
"xiaohongshu": "小红书",
}
def _extract_description(content: str, max_len: int = 200) -> str:
"""从 markdown 正文提取第一段有意义的文字作为 description"""
text = re.sub(r'^#\s+.*$', '', content, flags=re.MULTILINE)
text = re.sub(r'[#*>`~\[\]()\n]', ' ', text)
paragraphs = [p.strip() for p in text.split('\n\n') if p.strip()]
for p in paragraphs:
p = re.sub(r'\s+', ' ', p).strip()
if len(p) >= 15 and not p.startswith('http'):
return p[:max_len]
return content.replace('\n', ' ')[:max_len]
def _extract_tags_list(tags_html: str) -> list:
"""从 HTML tags 块提取纯标签列表"""
return re.findall(r'([^<]+)', tags_html)
def inject_geo_metadata(html: str, title: str, content: str, platform: str, tags_html: str = "") -> str:
"""向 HTML 注入 SEO/GEO 结构化元数据"""
description = _extract_description(content)
tags_list = _extract_tags_list(tags_html)
platform_name = PLATFORM_NAMES.get(platform, platform)
today = datetime.datetime.now().strftime("%Y-%m-%d")
# JSON-LD Article schema
json_ld = {
"@context": "https://schema.org",
"@type": "Article",
"headline": title,
"description": description,
"datePublished": today,
"dateModified": today,
"author": {
"@type": "Organization",
"name": "宇之然",
"url": "https://yu-zhi-ran.com"
},
"publisher": {
"@type": "Organization",
"name": "宇之然",
"url": "https://yu-zhi-ran.com"
},
"mainEntityOfPage": {
"@type": "WebPage",
"@id": f"https://yu-zhi-ran.com/article/{platform}"
},
}
if tags_list:
json_ld["keywords"] = ", ".join(tags_list[:8])
json_ld_str = json.dumps(json_ld, ensure_ascii=False)
meta_tags = f"""
"""
# 注入到 之前
html = html.replace("", meta_tags + "\n")
# 为 wechat + xiaohongshu 追加 Weibo/Wechat 兼容 meta
if platform in ("wechat", "xiaohongshu"):
html = html.replace("", """
""")
return html
class Writer:
def __init__(self, topic_id: str):
self.topic_id = topic_id
self.topic = self._load_topic()
outline_file = OUTLINE_DIR / TODAY / f"{topic_id}_outline.md"
if not outline_file.exists():
raise FileNotFoundError(f"Outline not found: {outline_file}")
self.outline_content = outline_file.read_text(encoding='utf-8')
research_file = DATA_DIR / "research" / TODAY / f"{topic_id}_research.md"
self.research_notes = research_file.read_text(encoding='utf-8') if research_file.exists() else ""
def _load_topic(self) -> Dict:
topic = get_topic_by_id(self.topic_id)
if not topic:
raise ValueError(f"Topic {self.topic_id} not found")
return topic
def _clean_title(self, title: str) -> str:
title = re.sub(r'[（(]约\s*\d+字[）)]', '', title)
title = re.sub(r'[（(]MVP[）)]', '', title)
title = re.sub(r'[（(][^）)]*?[）)]', '', title)
return title.strip()
def _parse_outline_sections(self) -> List[Dict]:
sections = []
current = None
for line in self.outline_content.splitlines():
if line.startswith("# "):
if current:
sections.append(current)
current = {"level": 1, "title": line[2:].strip(), "content": "", "section_type": "normal"}
elif line.startswith("## "):
if current:
sections.append(current)
title = line[3:].strip()
stype = "noise" if title in ("文章大纲", "大纲", "文章结构", "结构") else "normal"
current = {"level": 2, "title": title, "content": "", "section_type": stype}
elif line.startswith("### "):
if current:
sections.append(current)
current = {"level": 3, "title": line[4:].strip(), "content": "", "section_type": "normal"}
else:
if current and line.strip():
current['content'] = current.get('content', '') + line + "\n"
if current:
sections.append(current)
return sections
@staticmethod
def _clean_markdown(text: str) -> str:
return clean_markdown_content(text)
@staticmethod
def _is_outline_noise(line: str) -> bool:
stripped = line.strip()
if not stripped:
return True
if stripped.startswith('---'):
return True
if '大纲生成时间' in stripped:
return True
if stripped.startswith('*大纲'):
return True
if re.match(r'^\*{0,3}【[^】]*】\*{0,3}\s*$', stripped):
return True
return False
def _is_bullet_only(self, text: str) -> bool:
"""检查内容是否主要是要点列表（大纲格式），需要 LLM 展开"""
lines = [l.strip() for l in text.split('\n') if l.strip()]
if not lines:
return False
bullet_count = sum(1 for l in lines if l.startswith(('- ', '* ', '**', '+ ')))
return bullet_count / len(lines) > 0.4
def _expand_section(self, section: Dict, platform: str = "zhihu") -> str:
content = section.get('content', '').strip()
platform_prompt_key = f"section_expansion_{platform}"
if platform not in ("zhihu", "wechat", "xiaohongshu"):
platform_prompt_key = "section_expansion_zhihu"
if HAVE_LLM and content:
logger.info(f"LLM 扩写 [{platform}]: {section['title']} ({len(content)} chars)")
prompt = get_prompt(platform_prompt_key,
topic_title=self.topic['title'],
section_title=section['title'],
date=datetime.datetime.now().strftime('%Y年%m月%d日'),
content=content,
)
try:
params = get_prompt_params(platform_prompt_key) or {"temperature": 0.75, "max_tokens": 3000}
expanded = call_llm(prompt, temperature=params.get("temperature", 0.75), max_tokens=params.get("max_tokens", 3000))
if expanded:
cleaned = self._clean_markdown(expanded.strip())
if cleaned:
return cleaned
except Exception as e:
logger.warning(f"LLM 扩写失败 [{platform}]: {e}")
lines = [l.strip() for l in content.split('\n') if not self._is_outline_noise(l)]
if lines:
sentences = []
for line in lines:
text = line
for prefix in ['- ', '* ', '1. ', '2. ', '3. ', '4. ', '5. ']:
if line.startswith(prefix):
text = line[len(prefix):]
break
text = text.strip()
if text:
if text[-1] not in '。！？；':
text += '。'
sentences.append(text)
if sentences:
result = ' '.join(sentences)
return self._clean_markdown(result)
return ''
def generate_platform_markdown(self, platform: str = "zhihu") -> str:
sections = self._parse_outline_sections()
parts = []
for sec in sections:
if sec['level'] == 1:
if sec.get('content'):
expanded = self._expand_section(sec, platform)
if expanded:
parts.append(expanded + "\n")
continue
title_stripped = sec['title'].strip()
if title_stripped in ('文章大纲', '大纲', '文章结构', '结构'):
continue
if sec.get('section_type') == 'noise':
continue
heading = f"{'#' * sec['level']} {sec['title']}"
parts.append(heading)
if sec.get('content'):
expanded = self._expand_section(sec, platform)
parts.append(expanded + "\n")
full_md = "\n".join(parts).strip()
# 收集所有引用来源，统一添加到文末
refs = set()
for m in re.finditer(r'（来源：([^）]+)）', full_md):
refs.add(m.group(1).strip())
if refs:
# 去掉已有参考资料区，重新生成统一的
full_md = re.sub(r'\n---\n\*\*参考资料\*\*[\s\S]*$', '', full_md).strip()
ref_lines = "\n".join(f"- {r}" for r in sorted(refs))
full_md += f"\n\n---\n\n**参考资料**\n{ref_lines}"
return full_md
def _adapt_for_platform(self, markdown: str, platform: str) -> str:
cfg = PLATFORM_CONFIG[platform]
max_c = cfg['max_chars']
lines = markdown.split('\n')
if platform == "zhihu":
result = []
in_list = False
for line in lines:
stripped = line.strip()
# 数据类行 → 引用格式（知乎文章中引用数据能增强可信度）
if any(stripped.startswith(p) for p in ('据统计', '调研显示', '数据显示', '报告指出', '根据', '数据显示')):
line = f"> {line}"
# 列表保持原样（知乎支持 markdown 列表）
if stripped.startswith('- ') or stripped.startswith('* '):
if not in_list:
result.append('')
in_list = True
else:
in_list = False
result.append(line)
adapted = '\n'.join(result)
# 末尾加讨论引导（知乎算法权重：互动率）
if not any(kw in adapted for kw in ('你觉得', '你怎么看', '欢迎在评论区', '说说你的')):
adapted += "\n\n---\n\n你觉得这个观点有道理吗？你在工作中有没有类似的经验？欢迎在评论区聊聊。"
return adapted
if platform == "xiaohongshu":
result = []
char_count = 0
last_was_heading = False
for line in lines:
if char_count >= max_c:
break
stripped = line.strip()
if line.startswith('## '):
if not last_was_heading and result:
result.append('')
char_count += 1
line = f"## ✨ {line[3:]}"
last_was_heading = True
elif line.startswith('### '):
if not last_was_heading and result:
result.append('')
char_count += 1
line = f"### 💡 {line[4:]}"
last_was_heading = True
else:
last_was_heading = False
# 超长段落后拆行 + 每段前加点缀
if stripped and len(stripped) > 60:
sentences = [s.strip() for s in stripped.replace('。', '。\n').split('\n') if s.strip()]
for s in sentences:
if s and char_count < max_c:
result.append(s)
char_count += len(s)
continue
result.append(line)
char_count += len(line)
adapted = '\n'.join(result)
# 结尾加收藏引导（小红书算法权重：收藏率）
if '收藏' not in adapted:
adapted += "\n\n✨ 觉得有用的话点个收藏吧，下次需要的时候随时翻出来看~"
return adapted
if platform == "wechat":
result = []
for line in lines:
if line.startswith('### '):
result.append(f"\n**{line[4:]}**\n")
elif line.startswith('## '):
result.append(f"\n**{line[3:]}**\n")
elif line.strip() and len(line) > 80:
sentences = [s.strip() for s in line.replace('。', '。\n').split('\n') if s.strip()]
for s in sentences:
if s:
result.append(s)
else:
result.append(line)
adapted = '\n'.join(result)
return adapted
return markdown
def _get_platform_tags(self, platform: str) -> str:
field = self.topic.get('field', '')
title = self.topic.get('title', '')
core = self.topic.get('core_concept', '')
tag_prompts = {
"zhihu": get_prompt("tags_generation", platform="知乎", title=title, field=field, core=core),
"wechat": get_prompt("tags_generation", platform="公众号", title=title, field=field, core=core),
"xiaohongshu": get_prompt("tags_generation", platform="小红书", title=title, field=field, core=core),
}
if HAVE_LLM:
prompt = tag_prompts.get(platform, get_prompt("tags_generation", platform=platform, title=title, field=field, core=core))
try:
params = get_prompt_params("tags_generation")
tags_text = call_llm(prompt, temperature=params.get("temperature", 0.3), max_tokens=params.get("max_tokens", 500))
tags_text = strip_thinking(tags_text)
if tags_text:
tags = [t.strip('#') for t in tags_text.strip().split() if t.strip('#')]
if tags:
return " ".join(f'{t}' for t in tags[:5])
except Exception:
pass
tags = []
if field:
import re
parts = re.split(r'[/、与和及]', field)
for p in parts:
p = p.strip()
if len(p) >= 2:
tags.append(p)
if len(parts) == 1 and len(parts[0]) > 4:
for i in range(0, len(parts[0]), 2):
chunk = parts[0][i:i+2]
if len(chunk) == 2:
tags.append(chunk)
tags.pop(0)
platform_extra = {"zhihu": ["职场"], "xiaohongshu": ["生活"]}
for t in platform_extra.get(platform, []):
if t not in tags:
tags.append(t)
if not tags:
tags = ["科技"]
seen = set()
return " ".join(f'{t}' for t in tags if t not in seen and not seen.add(t))
def _optimize_title(self, platform: str) -> str:
original = self.topic['title']
if not HAVE_LLM:
return original
if platform == "zhihu":
prompt = get_prompt("title_optimize_zhihu",
title=original,
core=self.topic.get('core_concept', ''),
pain=self.topic.get('audience_pain', ''),
field=self.topic.get('field', ''),
)
elif platform == "wechat":
prompt = get_prompt("title_optimize_wechat",
title=original,
core=self.topic.get('core_concept', ''),
field=self.topic.get('field', ''),
)
elif platform == "xiaohongshu":
prompt = get_prompt("title_optimize_xhs",
title=original,
core=self.topic.get('core_concept', ''),
)
else:
prompt = f"给以下文章改个吸引人的{platform}标题：{original}"
try:
if platform in ("zhihu", "wechat", "xiaohongshu"):
params = get_prompt_params(f"title_optimize_{platform}")
resp = call_llm(prompt, temperature=params.get("temperature", 0.8), max_tokens=params.get("max_tokens", 1500))
else:
resp = call_llm(prompt, temperature=0.7)
resp = strip_thinking(resp)
titles = []
for line in resp.strip().split('\n'):
line = line.strip()
if not line:
continue
line = re.sub(r'^\d+[.、)\s]+', '', line)
line = line.strip('*#- \t"\'"''"')
# 跳过思考/建议类输出（如"不如："、"或者："、"建议方案"等）
if re.match(r'^(不如|或者|建议|推荐|参考|方案[一二三]|第[一二三]种|以[下是]|标题[一二三]|选项)', line):
continue
if line:
titles.append(line)
if titles:
best = titles[0][:80]
# 如果优化后标题与原文毫无关联或过短，回退原题
if len(best) < 4 or (len(set(best) & set(original)) < 2 and len(original) > 4):
logger.warning(f"标题优化结果异常「{best}」，回退原文")
return original
logger.info(f"标题优化 [{platform}]: {best}")
return best
except Exception as e:
logger.warning(f"标题优化失败: {e}")
return original
def generate_platform_html(self, markdown: str, platform: str) -> str:
title = self._optimize_title(platform)
adapted = self._adapt_for_platform(markdown, platform)
tpl_path = TEMPLATES_DIR / f"{platform}.html"
if tpl_path.exists():
template = tpl_path.read_text(encoding='utf-8')
else:
template = "{{TITLE}}{{TITLE}}
"
adapted = strip_ai_preface(adapted)
html = template.replace("{{TITLE}}", title).replace("{{DATE}}", TODAY).replace("{{GEN_TIME}}", GEN_TIME)
html_content = _md_parser(adapted)
# 仅插入头图（每个平台一篇一张，不过度）
html_content = insert_lead_image(
html_content, platform,
title=self.topic.get('title', title),
field=self.topic.get('field', ''),
)
html = html.replace("", html_content)
# 防御：清理可能在 LLM 输出中混入的 markdown 代码围栏和文件头
html = re.sub(r'^```+\w*\s*\n?', '', html)
html = re.sub(r'\n?```+\s*$', '', html)
html = html.strip()
tags_html = self._get_platform_tags(platform)
if tags_html:
html = html.replace("", tags_html)
else:
html = html.replace("", "")
html = inject_geo_metadata(html, title, adapted, platform, tags_html)
return html
def save_html(self, html: str, platform: str, *, title: str = "", content: str = "") -> str:
try:
save_article(self.topic_id, platform, html, title=title, content=content)
logger.info(f"文章写入数据库: {platform}_{self.topic_id}")
return f"db:{platform}_{self.topic_id}"
except Exception as e:
logger.warning(f"数据库保存失败: {e}")
return ""
def mark_draft(self):
update_topic_status(self.topic_id, 'review')
logger.info(f"选题 {self.topic_id} 状态已更新为待审查（数据库）")
def run(self):
logger.info("开始撰写阶段（三平台独立展开）")
results = {}
for platform in ["zhihu", "wechat", "xiaohongshu"]:
markdown = self.generate_platform_markdown(platform)
title = self._optimize_title(platform)
html = self.generate_platform_html(markdown, platform)
results[platform] = str(self.save_html(html, platform, title=title, content=markdown))
self.mark_draft()
logger.info(f"撰写完成，状态已更新为待审查")
return {"ok": True, "files": results}
def main():
import argparse
parser = argparse.ArgumentParser()
parser.add_argument('--topic-id', required=True, help='选题ID')
args = parser.parse_args()
w = Writer(args.topic_id)
result = w.run()
print(json.dumps(result, ensure_ascii=False))
sys.exit(0 if result['ok'] else 1)
if __name__ == "__main__":
main()