This commit is contained in:
Ubuntu
2026-07-03 16:58:28 +08:00
commit 32b43f85bd
6240 changed files with 1987678 additions and 0 deletions
@@ -0,0 +1,40 @@
# ============================================================
# 自动改写发布配置
# 修改后即时生效,无需重启任何服务
# ============================================================
# 时区(用于解析 schedule 中的时间)
timezone: "Asia/Shanghai"
# 发布时间表:每天哪些时刻发几篇
schedule:
- time: "08:00" # 早上 8 点发 1 篇
count: 1
- time: "20:00" # 晚上 8 点发 1 篇
count: 1
# 候选二级栏目(随机从中挑选)
# 如果某个栏目草稿用完,自动跳过换下一个
columns:
- 尿检
- 肾友
- 尿酸痛风
- 糖友
- 健康
# 挑选策略: random(随机) | round_robin(轮流)
pick_strategy: "random"
# ============================================================
# 发布模式 (publish_mode) — 仅控制公众号侧行为
# CMS 文章改写后统一标记为 published,不受此参数影响
# ============================================================
# "publish" — 直接发布(默认)
# 公众号草稿创建后自动提交发布 (draft/add → freepublish/submit)
# 文章在公众号立即可见
# "draft" — 存储草稿
# 仅创建公众号草稿 (draft/add),不自动提交发布
# 需登录微信公众平台手动点击发布,适合需要二次审核的场景
#
# 若要切换为草稿模式,将下行改为: publish_mode: "draft"
publish_mode: "draft"
@@ -0,0 +1,359 @@
#!/usr/bin/env python3
"""
自动定时改写发布调度器
由 cron 每分钟触发一次,读取 auto_config.yaml,判断当前时段是否需要发布。
如果当前时间落入某个 schedule slot 且该 slot 今日尚未执行,则随机从不同二级栏目挑选草稿,
调用 pipeline.run_pipeline() 改写并发布。
用法:
python3 auto_publish.py # 由 cron 调用
python3 auto_publish.py --dry-run # 模拟运行,不实际发布
python3 auto_publish.py --now # 忽略时间判断,立即执行一次
"""
import sys
import os
import json
import random
import logging
from datetime import datetime, date
from pathlib import Path
import yaml
# ---- 路径设置 ----
SCRIPT_DIR = os.path.dirname(os.path.abspath(__file__))
CONFIG_PATH = os.path.join(SCRIPT_DIR, "auto_config.yaml")
STATE_PATH = os.path.join(SCRIPT_DIR, "auto_state.json")
LOG_PATH = os.path.join(SCRIPT_DIR, "logs", "auto_publish.log")
sys.path.insert(0, SCRIPT_DIR)
from cms_ops import _get_conn, find_draft_by_column
from pipeline import run_pipeline
# ---- 日志 ----
os.makedirs(os.path.dirname(LOG_PATH), exist_ok=True)
logging.basicConfig(
level=logging.INFO,
format="%(asctime)s [%(levelname)s] %(message)s",
handlers=[
logging.FileHandler(LOG_PATH),
logging.StreamHandler(sys.stdout),
],
)
log = logging.getLogger("auto_publish")
# ============================================================
# 配置加载
# ============================================================
def load_config():
"""加载 YAML 配置"""
if not os.path.exists(CONFIG_PATH):
log.error(f"配置文件不存在: {CONFIG_PATH}")
sys.exit(1)
with open(CONFIG_PATH, "r") as f:
cfg = yaml.safe_load(f)
return cfg
# ============================================================
# 状态管理
# ============================================================
def load_state():
"""加载状态文件"""
if os.path.exists(STATE_PATH):
with open(STATE_PATH, "r") as f:
return json.load(f)
return {}
def save_state(state):
"""保存状态文件"""
with open(STATE_PATH, "w") as f:
json.dump(state, f, indent=2, ensure_ascii=False)
def today_str() -> str:
"""返回当天日期字符串,基于配置的时区"""
try:
cfg = load_config()
tz_name = cfg.get("timezone", "Asia/Shanghai")
from zoneinfo import ZoneInfo
tz = ZoneInfo(tz_name)
return datetime.now(tz).strftime("%Y-%m-%d")
except Exception:
return date.today().isoformat()
# ============================================================
# 文章选取
# ============================================================
def get_available_columns(cfg) -> list:
"""
返回有草稿的栏目列表。
每项: {"name": "尿检", "column_id": 19, "draft_count": 1703}
"""
conn = _get_conn()
cur = conn.cursor()
available = []
for col_name in cfg.get("columns", []):
cur.execute(
"""SELECT c.id, COUNT(a.id) as cnt
FROM columns c
LEFT JOIN articles a ON a.column_id = c.id AND a.status = 'draft'
WHERE c.name = ? AND c.parent_id IS NOT NULL
GROUP BY c.id""",
(col_name,),
)
row = cur.fetchone()
if row and row[1] > 0:
available.append({
"name": col_name,
"column_id": row[0],
"draft_count": row[1],
})
conn.close()
return available
def pick_articles(cfg, count: int, exclude_ids: set, used_columns_today: set = None, strategy_hint: str = None) -> list:
"""
从不同栏目中随机挑选 count 篇草稿,优先选择当天未用过的栏目。
Args:
cfg: 配置字典
count: 需要挑选的文章数
exclude_ids: 已发布过的文章 ID 集合(避免重复)
used_columns_today: 当天已用过的栏目名集合(避免重复栏目)
strategy_hint: 当前 strategy 提示(用于日志)
Returns:
[{"id": 123, "title": "...", "column": "尿检"}, ...]
"""
strategy = strategy_hint or cfg.get("pick_strategy", "random")
used_columns_today = used_columns_today or set()
available_cols = get_available_columns(cfg)
if not available_cols:
log.warning("所有栏目均无草稿可发")
return []
# 把栏目分成"未用过"和"已用过"两组,未用过的优先
fresh_cols = [c for c in available_cols if c["name"] not in used_columns_today]
used_cols = [c for c in available_cols if c["name"] in used_columns_today]
conn = _get_conn()
cur = conn.cursor()
picked = []
picked_columns = set()
# 按策略排列栏目顺序(未用过优先,已用过兜底)
if strategy == "random":
random.shuffle(fresh_cols)
random.shuffle(used_cols)
else:
state = load_state()
last_idx = state.get("last_column_index", 0)
fresh_cols = fresh_cols[last_idx:] + fresh_cols[:last_idx]
used_cols = used_cols[last_idx:] + used_cols[:last_idx]
shuffled = fresh_cols + used_cols
for _ in range(count):
found = False
for col in shuffled:
if col["name"] in picked_columns:
continue # 本次挑选内不重复栏目
cur.execute(
"""SELECT id, title FROM articles
WHERE column_id = ? AND status = 'draft'
AND id NOT IN ({})
ORDER BY created_at DESC LIMIT 1""".format(
",".join("?" * len(exclude_ids)) if exclude_ids else "0"
),
[col["column_id"]] + (list(exclude_ids) if exclude_ids else []),
)
row = cur.fetchone()
if row:
picked.append({
"id": row[0],
"title": row[1],
"column": col["name"],
})
picked_columns.add(col["name"])
exclude_ids.add(row[0])
found = True
break
if not found:
# 所有栏目都试过了,放宽"不同栏目"限制
for col in shuffled:
cur.execute(
"""SELECT id, title FROM articles
WHERE column_id = ? AND status = 'draft'
AND id NOT IN ({})
ORDER BY created_at DESC LIMIT 1""".format(
",".join("?" * len(exclude_ids)) if exclude_ids else "0"
),
[col["column_id"]] + (list(exclude_ids) if exclude_ids else []),
)
row = cur.fetchone()
if row:
picked.append({
"id": row[0],
"title": row[1],
"column": col["name"],
})
exclude_ids.add(row[0])
found = True
break
if not found:
log.warning("草稿不足,已无更多可选文章")
break
conn.close()
# 更新 round_robin 索引
if strategy == "round_robin" and picked:
last_col_name = picked[-1]["column"]
for i, col in enumerate(available_cols):
if col["name"] == last_col_name:
state = load_state()
state["last_column_index"] = (i + 1) % len(available_cols)
save_state(state)
break
return picked
# ============================================================
# 主逻辑
# ============================================================
def should_publish(cfg, current_dt) -> list:
"""
检查当前时间是否应该触发发布。
Returns:
应触发的 slot 列表,每项: {"time": "08:00", "count": 1}
"""
today = current_dt.strftime("%Y-%m-%d")
state = load_state()
# 不是同一天 → 重置状态
if state.get("date") != today:
state = {"date": today, "fulfilled_slots": [], "published_ids": [], "used_columns": []}
save_state(state)
fulfilled = set(state.get("fulfilled_slots", []))
due_slots = []
for slot in cfg.get("schedule", []):
slot_time = slot["time"] # "08:00"
if slot_time in fulfilled:
continue
# 当前时间 >= 计划时间?
slot_h, slot_m = map(int, slot_time.split(":"))
now_minutes = current_dt.hour * 60 + current_dt.minute
slot_minutes = slot_h * 60 + slot_m
if now_minutes >= slot_minutes:
due_slots.append(slot)
return due_slots
def execute_slot(cfg, slot, state, dry_run=False):
"""
执行一个时间槽的发布任务。
Returns:
成功发布的文章数
"""
count = slot.get("count", 1)
slot_key = slot["time"]
log.info(f"⏰ 时段 {slot_key} 触发,计划发布 {count} 篇")
exclude_ids = set(state.get("published_ids", []))
used_columns_today = set(state.get("used_columns", []))
articles = pick_articles(cfg, count, exclude_ids, used_columns_today)
if not articles:
log.warning(f"时段 {slot_key}: 无可选草稿,跳过")
return 0
success = 0
for art in articles:
log.info(f"📝 开始处理 [{art['column']}] {art['title']} (id={art['id']})")
if dry_run:
log.info(f" [DRY-RUN] 将调用 run_pipeline({art['id']})")
success += 1
state.setdefault("published_ids", []).append(art["id"])
state.setdefault("used_columns", []).append(art["column"])
else:
try:
run_pipeline(art["id"])
log.info(f" ✅ 发布成功: {art['title']}")
success += 1
state.setdefault("published_ids", []).append(art["id"])
state.setdefault("used_columns", []).append(art["column"])
except Exception as e:
log.error(f" ❌ 发布失败: {art['title']} — {e}")
# 标记 slot 完成
if success > 0 or not articles:
state.setdefault("fulfilled_slots", []).append(slot_key)
save_state(state)
log.info(f"时段 {slot_key} 完成: {success}/{count} 篇成功")
return success
def main():
dry_run = "--dry-run" in sys.argv
force_now = "--now" in sys.argv
cfg = load_config()
# 获取当前时间(按配置时区)
tz_name = cfg.get("timezone", "Asia/Shanghai")
try:
from zoneinfo import ZoneInfo
tz = ZoneInfo(tz_name)
except Exception:
tz = None
current_dt = datetime.now(tz) if tz else datetime.now()
if force_now:
log.info(f"⚡ 手动执行所有未完成 slot [{current_dt.strftime('%Y-%m-%d %H:%M:%S')}]")
state = load_state()
today = current_dt.strftime("%Y-%m-%d")
if state.get("date") != today:
state = {"date": today, "fulfilled_slots": [], "published_ids": [], "used_columns": []}
for slot in cfg.get("schedule", []):
if slot["time"] not in state.get("fulfilled_slots", []):
execute_slot(cfg, slot, state, dry_run=dry_run)
return
due_slots = should_publish(cfg, current_dt)
if not due_slots:
return # 静默退出,cron 定时触发的正常情况
state = load_state()
for slot in due_slots:
execute_slot(cfg, slot, state, dry_run=dry_run)
if __name__ == "__main__":
main()
@@ -0,0 +1,12 @@
{
"date": "2026-06-22",
"fulfilled_slots": [
"08:00"
],
"published_ids": [
4011
],
"used_columns": [
"糖友"
]
}
@@ -0,0 +1,258 @@
"""
CMS 数据库操作工具 — 草稿文章查找、内容更新
用法:
from cms_ops import find_draft_by_title, update_article_content
"""
import sqlite3
import os
from datetime import datetime
from typing import Optional, List, Dict, Any
# 数据库路径(相对脚本所在目录)
SCRIPT_DIR = os.path.dirname(os.path.abspath(__file__))
PROJECT_ROOT = os.path.dirname(SCRIPT_DIR) # website-new 根目录
DB_PATH = os.path.join(PROJECT_ROOT, "backend", "cms.db")
def _get_conn():
"""获取数据库连接"""
conn = sqlite3.connect(DB_PATH)
conn.row_factory = sqlite3.Row
return conn
def find_draft_by_title(keyword: str) -> Optional[Dict[str, Any]]:
"""
按标题关键词查找最新一篇草稿文章。
Args:
keyword: 标题关键词
Returns:
文章字典,未找到返回 None
"""
conn = _get_conn()
cursor = conn.cursor()
cursor.execute(
"""SELECT * FROM articles
WHERE status = 'draft' AND title LIKE ?
ORDER BY created_at DESC
LIMIT 1""",
(f"%{keyword}%",),
)
row = cursor.fetchone()
conn.close()
if row:
return dict(row)
return None
def find_draft_by_column(column_name: str) -> Optional[Dict[str, Any]]:
"""
查找某栏目下最新一篇草稿文章。
Args:
column_name: 栏目名称(模糊匹配)
Returns:
文章字典,未找到返回 None
"""
conn = _get_conn()
cursor = conn.cursor()
cursor.execute(
"""SELECT articles.* FROM articles
JOIN columns ON articles.column_id = columns.id
WHERE articles.status = 'draft' AND columns.name LIKE ?
ORDER BY articles.created_at DESC
LIMIT 1""",
(f"%{column_name}%",),
)
row = cursor.fetchone()
conn.close()
if row:
return dict(row)
return None
def find_draft_by_id(article_id: int) -> Optional[Dict[str, Any]]:
"""
按 ID 查找草稿文章。
Args:
article_id: 文章 ID
Returns:
文章字典,未找到返回 None
"""
conn = _get_conn()
cursor = conn.cursor()
cursor.execute(
"SELECT * FROM articles WHERE id = ? AND status = 'draft'",
(article_id,),
)
row = cursor.fetchone()
conn.close()
if row:
return dict(row)
return None
def list_latest_drafts(limit: int = 10) -> List[Dict[str, Any]]:
"""
列出最新的 N 篇草稿文章。
Args:
limit: 返回数量
Returns:
文章字典列表
"""
conn = _get_conn()
cursor = conn.cursor()
cursor.execute(
"""SELECT id, title, column_id, content_type,
length(content) as content_len, created_at
FROM articles
WHERE status = 'draft'
ORDER BY created_at DESC
LIMIT ?""",
(limit,),
)
rows = cursor.fetchall()
conn.close()
return [dict(r) for r in rows]
def get_article_full(article_id: int) -> Optional[Dict[str, Any]]:
"""
获取文章完整内容(不限制状态)。
Args:
article_id: 文章 ID
Returns:
文章字典,包含 content 字段
"""
conn = _get_conn()
cursor = conn.cursor()
cursor.execute("SELECT * FROM articles WHERE id = ?", (article_id,))
row = cursor.fetchone()
conn.close()
if row:
return dict(row)
return None
def update_article_content(
article_id: int,
new_content: str,
status: str = "published",
new_title: Optional[str] = None,
new_author: Optional[str] = None,
new_cover_image: Optional[str] = None,
) -> bool:
"""
更新文章内容、状态和可选的标题/作者/封面图。
Args:
article_id: 文章 ID
new_content: 改写后的正文
status: 新状态,默认 'published'
new_title: 可选,改写后的标题
new_author: 可选,改写后的作者
new_cover_image: 可选,新的封面图路径
Returns:
是否更新成功
"""
conn = _get_conn()
cursor = conn.cursor()
now = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
sets = ["content = ?", "status = ?", "updated_at = ?"]
params = [new_content, status, now]
if new_title:
sets.append("title = ?")
params.append(new_title)
if new_author:
sets.append("author = ?")
params.append(new_author)
if new_cover_image:
sets.append("cover_image = ?")
params.append(new_cover_image)
params.append(article_id)
cursor.execute(
f"UPDATE articles SET {', '.join(sets)} WHERE id = ?",
params,
)
conn.commit()
affected = cursor.rowcount
conn.close()
return affected > 0
def count_drafts() -> int:
"""统计草稿文章数量"""
conn = _get_conn()
cursor = conn.cursor()
cursor.execute("SELECT COUNT(*) FROM articles WHERE status = 'draft'")
count = cursor.fetchone()[0]
conn.close()
return count
# ---- 命令行模式 ----
if __name__ == "__main__":
import sys
if len(sys.argv) < 2:
print("用法:")
print(" python cms_ops.py list [limit] 列表最新草稿")
print(" python cms_ops.py search <keyword> 按标题搜索")
print(" python cms_ops.py get <id> 获取文章完整内容")
print(" python cms_ops.py count 统计草稿数量")
sys.exit(0)
cmd = sys.argv[1]
if cmd == "list":
limit = int(sys.argv[2]) if len(sys.argv) > 2 else 10
drafts = list_latest_drafts(limit)
print(f"最新 {len(drafts)} 篇草稿:\n")
for d in drafts:
title = d["title"][:50]
print(f" ID={d['id']:<5} | {title:<50s} | {d['content_len']} chars | {d['created_at']}")
elif cmd == "search":
keyword = sys.argv[2]
article = find_draft_by_title(keyword)
if article:
print(f"找到: ID={article['id']} | {article['title']}")
print(f"内容类型: {article['content_type']}")
print(f"内容长度: {len(article.get('content', ''))} chars")
print(f"摘要: {article.get('summary', '')}")
print(f"\n正文:\n{article.get('content', '')[:500]}...")
else:
print(f"未找到匹配 '{keyword}' 的草稿")
elif cmd == "get":
article_id = int(sys.argv[2])
article = get_article_full(article_id)
if article:
print(f"ID={article['id']} | {article['title']}")
print(f"状态: {article['status']}")
print(f"内容类型: {article['content_type']}")
print(f"封面: {article.get('cover_image', '')}")
print(f"\n正文:\n{article.get('content', '')}")
else:
print(f"找不到 ID={article_id} 的文章")
elif cmd == "count":
print(f"草稿文章: {count_drafts()} 篇")
@@ -0,0 +1,962 @@
#!/usr/bin/env python3
"""
CMS 改写 → 公众号发布 一键流水线
两种模式:
1. 自动改写模式 (默认): python pipeline.py <article_id>
自己调 DeepSeek LLM 改写,然后发布
2. 传入改写内容模式: python pipeline.py <article_id> --content-file <markdown文件>
用预改写好的内容直接发布(志玲模式)
环境:
依赖 openai 库: pip install openai
API Key 从 ~/.nanobot/config.json 读取 DeepSeek 配置
"""
import sys
import os
import re
import json
import time
import random
import argparse
import yaml
from typing import Optional, Dict, Any
# 确保脚本在 rewrite/ 目录执行也能正常工作
SCRIPT_DIR = os.path.dirname(os.path.abspath(__file__))
PROJECT_ROOT = os.path.dirname(SCRIPT_DIR)
AUTO_CONFIG_PATH = os.path.join(SCRIPT_DIR, "auto_config.yaml")
BAOYU_DIR = "/home/ubuntu/.nanobot/workspace/skills/baoyu-post-to-wechat"
sys.path.insert(0, BAOYU_DIR)
sys.path.insert(0, SCRIPT_DIR)
from cms_ops import get_article_full, update_article_content
from baoyu_optimized import WeChatPublisher
# ---------- 配置 ----------
MODEL = "deepseek-v4-flash"
NANOBOT_CONFIG = os.path.expanduser("~/.nanobot/config.json")
SYSTEM_PROMPT = """你是一个专业的医疗健康内容编辑。请将下面的文章改写为适合微信公众号发布的版本。
改写要求:
1. **第一行必须是标题**: 以 `# ` 开头。⚠️ **必须重新创作标题,严禁照搬原标题原句**。可以换表述、调语序、换措辞,但必须保持原文核心含义不变(不曲解、不偏离主旨)。标题中禁止使用任何emoji表情符号
2. **保持医疗准确性**: 不得改变任何医学事实、数据、诊断标准、用药建议
3. **风格口语化**: 用亲近自然的语气,像医生朋友在聊天。正文适当使用emoji(每段不超过1个),但标题禁止emoji
4. **保留结构**: 保留原文的章节标题和信息层次,但标题可以重新提炼得更吸引人
5. **精简优化**: 去除重复表述,合并相近内容,让文章更紧凑
6. **清理无关内容**: 去掉原文作者署名、转载来源、第三方推广链接(如"相关阅读"、"更多精彩内容"、"推荐阅读"、"参考文献"等板块)
7. **适合手机阅读**: 段落短小(2-3句一段),善用小标题和列表
8. **只输出改写后的 markdown**,不要加任何解释说明
输出格式:
- 第一行: `# 改写后的标题`
- 正文: 纯 markdown,可以包含 **粗体**、无序列表 -、## 二级标题、段间空行分隔"""
# ---------- 图片生成配置 ----------
IMAGE_GEN_ENABLED = True # 是否启用配图生成
IMAGE_GEN_COUNT = 1 # 目标生成图片数量
IMAGE_STORAGE_DIR = os.path.join(PROJECT_ROOT, "storage", "admin", "images")
BANNERS_DIR = os.path.join(PROJECT_ROOT, "storage", "admin", "images", "banners")
ADS_IMAGE = os.path.join(PROJECT_ROOT, "storage", "admin", "images", "ads", "0.jpg")
IMAGE_PROMPT_SYSTEM = """你是一个图片提示词专家。根据下面给出的文章片段,生成一个适合配图的简短中文描述(用于AI生图),要求:
1. 描述要简洁具体,20-40个字
2. 风格要求:温暖、专业、治愈的视觉风格
3. 内容要与上下文紧密相关
4. 只输出描述文字,不要加任何解释或前缀"""
# 混元生图 (TokenHub MAS API)
HUNYUAN_API_KEY = "sk-qgPNJwuOGji7I7ixqNbbkWaTIvF1MrYTRWQnEZ3ZRVU4u7y3"
HUNYUAN_API_BASE = "https://tokenhub.tencentmaas.com/v1/api/image"
HUNYUAN_MODEL = "hy-image-v3.0"
# Seedream(降级)
SEEDREAM_API_KEY = "83a29955-c633-449f-97f1-f49b62295cf6"
SEEDREAM_BASE_URL = "https://ark.cn-beijing.volces.com/api/v3"
SEEDREAM_MODEL = "doubao-seedream-4-5-251128"
# ---------- 工具函数 ----------
def load_deepseek_key() -> str:
"""从 nanobot 配置读取 DeepSeek API Key"""
try:
with open(NANOBOT_CONFIG) as f:
config = json.load(f)
key = config["providers"]["deepseek"]["apiKey"]
if not key:
raise ValueError("DeepSeek API key 为空")
return key
except Exception as e:
print(f"❌ 读取 DeepSeek API Key 失败: {e}", file=sys.stderr)
print(f" 配置文件: {NANOBOT_CONFIG}", file=sys.stderr)
sys.exit(1)
def call_deepseek(markdown: str, api_key: str) -> str:
"""调用 DeepSeek API 改写文章"""
try:
from openai import OpenAI
except ImportError:
print("❌ 需要安装 openai: pip install openai", file=sys.stderr)
sys.exit(1)
client = OpenAI(api_key=api_key, base_url="https://api.deepseek.com")
print("⏳ 调 DeepSeek 改写中...", file=sys.stderr)
response = client.chat.completions.create(
model=MODEL,
messages=[
{"role": "system", "content": SYSTEM_PROMPT},
{"role": "user", "content": markdown},
],
temperature=0.7,
max_tokens=4096,
)
result = response.choices[0].message.content.strip()
tokens = response.usage.total_tokens if response.usage else "?"
print(f" 消耗 {tokens} tokens", file=sys.stderr)
return result
def resolve_cover_path(cover_url: Optional[str]) -> Optional[str]:
"""将 CMS 封面 URL 转为本地绝对路径"""
if not cover_url:
return None
# cover_url 格式: /storage/admin/images/202605/xxx.jpg
path = os.path.join(PROJECT_ROOT, "..", cover_url.lstrip("/"))
# 某些情况下 cover_url 可能不带前导 /
alt_path = os.path.join(PROJECT_ROOT, "storage", "admin", "images",
*cover_url.split("/")[-2:])
for p in [path, alt_path]:
if os.path.exists(p):
return os.path.normpath(p)
print(f"⚠️ 封面图未找到: {cover_url}", file=sys.stderr)
return None
# ---------- 标题提取 ----------
def _extract_title(markdown: str) -> tuple[str, str]:
"""从 markdown 首行提取 `# 标题`,返回 (标题, 去除标题后的正文)"""
lines = markdown.split("\n")
for i, line in enumerate(lines):
stripped = line.strip()
if stripped.startswith("# ") and not stripped.startswith("## "):
title = stripped[2:].strip()
# 去掉标题行,保留后续内容
remaining = "\n".join(lines[i + 1:]).strip()
return title, remaining
# 没有找到标题,返回空标题 + 原文
return "", markdown
# ---------- emoji 清理 ----------
def _strip_emoji(text: str) -> str:
"""移除文本中的 emoji 表情符号,用于清理标题。
策略:先移除非 BMP 字符(覆盖绝大多数现代 emoji),
再清理少量 BMP 区 emoji 符号。
保留 ⚠️、①②、℃ 等实用符号不被误删。
"""
# 1. 移除非 BMP 字符(Supplementary Plane,覆盖 😀🚀🎯 等几乎全部多彩 emoji)
text = re.sub(r'[^\u0000-\uFFFF]+', '', text)
# 2. 移除 BMP 区的 emoji/装饰符号
emoji_pattern = re.compile(
"["
"\U00002702-\U000027B0" # Dingbats (✂-➰)
"\U00002194-\U00002199" # Arrows (↔↕↖↗↘↙)
"\U00002B05-\U00002B07" # More arrows (⬅⬆⬇)
"\U00002934-\U00002935" # Curved arrows (⤴⤵)
"\U000025AA-\U000025AB" # Small squares (▪▫)
"\U000025B6" # Play ▶
"\U000025C0" # Reverse ◀
"\U000025FB-\U000025FE" # Medium/white squares (◻◼◽◾)
"\U00002614-\U00002615" # Umbrella/Rain (☔☕)
"\U00002648-\U00002653" # Zodiac etc (♈-♓)
"\U000026AA-\U000026AB" # Circles (⚪⚫)
"\U000026BD-\U000026BE" # Sports (⚽⚾)
"\U000026C4-\U000026C5" # Snowman/No clouds (⛄⛅)
"\U000026CE" # Ophiuchus (⛎)
"\U000026D4" # No entry (⛔)
"\U000026EA" # Church (⛪)
"\U000026F2-\U000026F3" # Fountain/Flag (⛲⛳)
"\U000026F5" # Sailboat (⛵)
"\U000026FA" # Tent (⛺)
"\U000026FD" # Fuel pump (⛽)
"\U00002714" # Checkmark (✔)
"\U00002716" # Cross (✖)
"\U00002728" # Sparkles (✨)
"\U00002733-\U00002734" # Star-tip (✳✴)
"\U00002744-\U00002745" # Snowflake (❄❅)
"\U00002747" # Sparkle (❇)
"\U00002757" # Cross mark (❌)
"\U00002763-\U00002764" # Heart (❣❤)
"\U00002795-\U00002797" # Plus/Minus/Divide (➕➖➗)
"\U00002B1B-\U00002B1C" # Large squares (⬛⬜)
"\U00002B50" # Star ⭐
"\U00002B55" # Circle ⭕
"\U00003030" # Wavy dash 〰
"\U0000303D" # Part alternation mark 〽
"\U00003297" # Congrats ㊗
"\U00003299" # Secret ㊙
"\U0000FE00-\U0000FE0F" # Variation Selectors
"\U0000200D" # ZWJ
"]+", flags=re.UNICODE)
return emoji_pattern.sub('', text).strip()
# ---------- 封面图 ----------
def _random_banner() -> Optional[str]:
"""从 banners 目录随机选一张图片,返回绝对路径"""
if not os.path.isdir(BANNERS_DIR):
return None
images = [f for f in os.listdir(BANNERS_DIR)
if f.lower().endswith(('.jpg', '.jpeg', '.png'))]
if not images:
return None
chosen = random.choice(images)
path = os.path.join(BANNERS_DIR, chosen)
print(f" 🎲 随机封面: {chosen}", file=sys.stderr)
return path
# ---------- 内容清理 ----------
def _clean_content(markdown: str) -> str:
"""清理第三方链接、作者署名等无关内容"""
lines = markdown.split("\n")
cleaned = []
skip_until_section_end = False
for line in lines:
stripped = line.strip().lower()
# 检测需要跳过的板块标题
if re.match(r'^#{1,3}\s*(相关阅读|更多精彩|推荐阅读|延伸阅读|参考文献|参考资料|热门推荐|往期回顾|猜你喜欢|广告)', stripped):
skip_until_section_end = True
continue
# 板块结束(遇到新标题 或 空行不再累加)
if skip_until_section_end:
if stripped.startswith("#") and not re.match(r'^#{1,3}\s*(相关阅读|更多精彩|推荐阅读|延伸阅读|参考文献|参考资料|热门推荐|往期回顾|猜你喜欢|广告)', stripped):
skip_until_section_end = False
cleaned.append(line)
continue
# 跳过带来源/作者标记的行(行首)
if re.match(r'^(原文作者|作者[::]|来源[::]|转载自|文[//]\s*|编辑[::]|排版[::]|审核[::])', stripped):
continue
# 跳过含作者+格式的元数据行(如 "尿酸痛风 作者: 刘松 格式: Markdown")
if re.search(r'作者[::]\s*\S+', stripped) and re.search(r'格式[::]', stripped):
continue
# 跳过独立的格式声明行
if re.match(r'^格式[::]\s*(Markdown|HTML|TXT|TEXT)', stripped):
continue
# 跳过明显的第三方链接(纯 URL 行)
if re.match(r'^https?://', stripped) and 'mp.weixin' not in stripped and 'niaodaifu' not in stripped:
continue
cleaned.append(line)
return "\n".join(cleaned)
# ---------- 内容质量检查 ----------
# LLM 占位回复检测模式(当原文无文字时 DeepSeek 会返回这类消息)
_LLM_PLACEHOLDER_PATTERNS = [
r'原文内容(全部|仅|只).*(图片|链接|无法显示)',
r'没有可改写的文本',
r'请提供.*文字',
r'请.*提供.*文章.*内容',
r'一旦获得文字内容',
r'您好,您提供的.*(图片|链接)',
r'为了生成适合',
r'无法生成适合',
r'请确保文章包含文字',
]
_MIN_TEXT_LENGTH = 200 # 原文纯文本低于此值视为无实质内容
def _strip_images(markdown: str) -> str:
"""移除 markdown 图片语法,返回纯文本"""
return re.sub(r'!\[.*?\]\([^)]+\)', '', markdown)
def _check_original_content(content: str) -> int:
"""
检查原文是否有足够的文字内容。
返回纯文本字符数;低于 _MIN_TEXT_LENGTH 抛出 ValueError。
"""
text = _strip_images(content)
# 也移除 markdown 链接、HTML 标签等
text = re.sub(r'\[([^\]]*)\]\([^)]+\)', r'\1', text) # 链接文字保留
text = re.sub(r'<[^>]+>', '', text) # HTML 标签
text = re.sub(r'\s+', ' ', text).strip()
text_len = len(text)
if text_len < _MIN_TEXT_LENGTH:
raise ValueError(
f"原文文字内容不足({text_len} 字符 < {_MIN_TEXT_LENGTH}),"
f"可能全部为图片,跳过改写"
)
return text_len
def _check_llm_response(content: str) -> None:
"""
检查 LLM 返回是否是无意义的占位回复。
如果是占位回复则抛出 ValueError。
"""
for pattern in _LLM_PLACEHOLDER_PATTERNS:
if re.search(pattern, content):
raise ValueError(
f"LLM 返回疑似占位回复(匹配模式: {pattern}),不发布"
)
# 额外检查:改写后内容过短(< 200 字)也视为无效
if len(content) < _MIN_TEXT_LENGTH:
raise ValueError(
f"改写后内容过短({len(content)} 字符 < {_MIN_TEXT_LENGTH}),不发布"
)
# ---------- 正文末尾广告图 ----------
def _append_ads_image(markdown: str) -> str:
"""在文章末尾添加固定广告图"""
if not os.path.exists(ADS_IMAGE):
print(f" ⚠️ 广告图不存在: {ADS_IMAGE}", file=sys.stderr)
return markdown
web_path = "/storage/admin/images/ads/0.jpg"
print(f" 📢 追加广告图: {web_path}", file=sys.stderr)
return markdown.rstrip() + f"\n\n![广告]({web_path})\n"
# ---------- 图片生成 ----------
def _make_image_prompt(context_text: str, api_key: str) -> str:
"""用 LLM 从文章片段生成图片 prompt"""
try:
from openai import OpenAI
except ImportError:
print("⚠️ openai 未安装,使用默认图片 prompt", file=sys.stderr)
return "温暖的医疗健康主题插图,简洁专业的扁平设计风格"
client = OpenAI(api_key=api_key, base_url="https://api.deepseek.com")
try:
response = client.chat.completions.create(
model=MODEL,
messages=[
{"role": "system", "content": IMAGE_PROMPT_SYSTEM},
{"role": "user", "content": context_text},
],
temperature=0.3,
max_tokens=200,
)
prompt = response.choices[0].message.content.strip()
print(f" 🎨 图片 prompt: {prompt}", file=sys.stderr)
return prompt
except Exception as e:
print(f" ⚠️ 图片 prompt 生成失败: {e}", file=sys.stderr)
return "温暖的医疗健康主题插图,简洁专业的扁平设计风格"
def _generate_image(prompt: str, output_path: str,
resolution: str = "1024:1024") -> bool:
"""调用混元生图生成图片,失败则降级到 Seedream"""
success = _hunyuan_image(prompt, output_path, resolution)
if success:
return True
print(" ↪ 降级到 Seedream...", file=sys.stderr)
return _seedream_image(prompt, output_path, resolution)
def _hunyuan_image(prompt: str, output_path: str,
resolution: str = "1024:1024") -> bool:
"""腾讯混元生图 hy-image-v3.0 (TokenHub MAS API)
异步提交流程: submit → poll query → download
"""
import requests
if not HUNYUAN_API_KEY:
return False
try:
# Step 1: 提交生图任务
submit_url = f"{HUNYUAN_API_BASE}/submit"
submit_payload = {
"model": HUNYUAN_MODEL,
"prompt": prompt,
"logo_add": 0, # 不添加"AI生成"水印
}
# 可选参数:尺寸
parts = resolution.replace(":", "x").split("x")
if len(parts) == 2:
try:
w, h = int(parts[0]), int(parts[1])
submit_payload["width"] = w
submit_payload["height"] = h
except ValueError:
pass
headers = {
"Authorization": f"Bearer {HUNYUAN_API_KEY}",
"Content-Type": "application/json",
}
resp = requests.post(submit_url, json=submit_payload, headers=headers, timeout=30)
resp.raise_for_status()
result = resp.json()
task_id = result.get("id")
if not task_id:
print(f" ⚠️ 混元提交失败: {result}", file=sys.stderr)
return False
print(f" 📤 混元任务已提交: {task_id}", file=sys.stderr)
# Step 2: 轮询查询结果
query_url = f"{HUNYUAN_API_BASE}/query"
query_payload = {"model": HUNYUAN_MODEL, "id": task_id}
for _ in range(30): # 最多等 150 秒
time.sleep(5)
qr = requests.post(query_url, json=query_payload, headers=headers, timeout=30)
qr.raise_for_status()
qresult = qr.json()
status = qresult.get("status", "")
if status == "completed":
# 提取图片 URL (Tencent MAS 返回 data 数组)
data = qresult.get("data", [])
if not data:
print(f" ⚠️ 混元无图片: {qresult}", file=sys.stderr)
return False
img_obj = data[0]
img_url = img_obj.get("url", "")
img_b64 = img_obj.get("b64_json", "")
if img_url:
r = requests.get(img_url, timeout=60)
r.raise_for_status()
img_data = r.content
elif img_b64:
import base64
img_data = base64.b64decode(img_b64)
else:
print(f" ⚠️ 混元无可用图片格式: {img_obj}", file=sys.stderr)
return False
os.makedirs(os.path.dirname(output_path) or ".", exist_ok=True)
with open(output_path, "wb") as f:
f.write(img_data)
print(f" ✅ 混元生图: {os.path.basename(output_path)} ({len(img_data)} bytes)", file=sys.stderr)
return True
elif status in ("failed", "cancelled"):
print(f" ⚠️ 混元任务失败: status={status}, {qresult}", file=sys.stderr)
return False
# 其他状态 (processing, queued, etc.) — 继续等待
print(f" ⏳ 混元: {status}", file=sys.stderr)
print(f" ⚠️ 混元超时 (150s)", file=sys.stderr)
return False
except Exception as e:
print(f" ⚠️ 混元异常: {e}", file=sys.stderr)
return False
def _seedream_image(prompt: str, output_path: str,
resolution: str = "1024:1024") -> bool:
"""Seedream (火山引擎 ARK)"""
import requests
# Seedream 要求 ≥ 3,686,400 像素,自动放大
parts = resolution.replace(":", "x").split("x")
if len(parts) == 2:
try:
w, h = int(parts[0]), int(parts[1])
if w * h < 3686400:
scale = (3686400 / (w * h)) ** 0.5
w2, h2 = (int(w * scale) // 16) * 16, (int(h * scale) // 16) * 16
resolution = f"{w2}x{h2}"
except ValueError:
pass
try:
url = f"{SEEDREAM_BASE_URL}/images/generations"
resp = requests.post(
url,
headers={
"Authorization": f"Bearer {SEEDREAM_API_KEY}",
"Content-Type": "application/json",
},
json={
"model": SEEDREAM_MODEL,
"prompt": prompt,
"size": resolution,
"n": 1,
},
timeout=120,
)
resp.raise_for_status()
data = resp.json()
image_url = data["data"][0].get("url")
if not image_url:
return False
r = requests.get(image_url, timeout=60)
r.raise_for_status()
os.makedirs(os.path.dirname(output_path) or ".", exist_ok=True)
with open(output_path, "wb") as f:
f.write(r.content)
print(f" ✅ Seedream: {os.path.basename(output_path)} ({len(r.content)} bytes)", file=sys.stderr)
return True
except Exception as e:
print(f" ⚠️ Seedream: {e}", file=sys.stderr)
return False
def _remove_watermark(image_path: str) -> None:
"""去除混元生图右下角'AI生成'水印:OpenCV修复 + 背景覆盖 + 官网logo叠加"""
try:
import cv2
import numpy as np
from PIL import Image, ImageDraw
LOGO_PATH = os.path.join(
PROJECT_ROOT,
"storage/admin/images/setting/20260506_100405_89971b62.png",
)
if not os.path.exists(LOGO_PATH):
return
img_cv = cv2.imread(image_path)
if img_cv is None:
return
h, w = img_cv.shape[:2]
# ---- 第1步:检测水印区域并修复 ----
# 右下角扫描: 30%宽 × 25%高(扩大扫描范围以捕获完整水印)
scan_w = int(w * 0.30)
scan_h = int(h * 0.25)
corner = img_cv[h - scan_h : h, w - scan_w : w]
# 转灰度,用大核高斯模糊做背景估计,diff 找出局部异常亮区
gray = cv2.cvtColor(corner, cv2.COLOR_BGR2GRAY)
blurred = cv2.GaussianBlur(gray, (51, 51), 0)
diff = cv2.subtract(gray, blurred)
# 水印文字比周围亮,自适应阈值
diff_max = diff.max()
threshold = max(8, int(diff_max * 0.35))
_, wm_mask = cv2.threshold(diff, threshold, 255, cv2.THRESH_BINARY)
# 膨胀连接文字笔画,扩大修复区域
kernel = np.ones((5, 5), np.uint8)
wm_mask = cv2.dilate(wm_mask, kernel, iterations=2)
# 补充右下角最小覆盖区(兜底:混元水印始终在右下角)
fixed_mask = np.zeros((scan_h, scan_w), dtype=np.uint8)
fixed_mask[int(scan_h * 0.4):, int(scan_w * 0.6):] = 255
wm_mask = cv2.bitwise_or(wm_mask, fixed_mask)
wm_pixels = wm_mask.sum() // 255
# 记录水印在扫描区中的边界框(全图坐标)
ys, xs = np.where(wm_mask > 0)
wm_left = w - scan_w + xs.min()
wm_top = h - scan_h + ys.min()
wm_right = w - scan_w + xs.max()
wm_bottom = h - scan_h + ys.max()
if wm_pixels > 200:
# 扩展到全图做修复(加大 inpaint 半径)
full_mask = np.zeros((h, w), dtype=np.uint8)
full_mask[h - scan_h : h, w - scan_w : w] = wm_mask
img_cv = cv2.inpaint(img_cv, full_mask, 5, cv2.INPAINT_TELEA)
print(f" 🎨 水印修复完成 ({wm_pixels} 像素)", file=sys.stderr)
else:
# 未检测到水印时,使用默认右下角区域
wm_left = int(w * 0.70)
wm_top = int(h * 0.75)
wm_right = w
wm_bottom = h
print(f" ℹ️ 未检测到明确水印,使用默认右下角覆盖", file=sys.stderr)
cv2.imwrite(image_path, img_cv, [cv2.IMWRITE_JPEG_QUALITY, 95])
# ---- 第2步:背景覆盖 + 叠加官网logo ----
img_pil = Image.open(image_path).convert("RGBA")
logo = Image.open(LOGO_PATH).convert("RGBA")
# 计算覆盖区域:从水印左上角延伸到图片右下角,加 padding
padding = int(min(w, h) * 0.015)
patch_left = max(0, wm_left - padding)
patch_top = max(0, wm_top - padding)
patch_right = w # 延伸到右边缘
patch_bottom = h # 延伸到底边缘
patch_w = patch_right - patch_left
patch_h = patch_bottom - patch_top
# 创建白色圆角背景(高不透明度确保完全覆盖水印)
patch = Image.new("RGBA", (patch_w, patch_h), (0, 0, 0, 0))
draw = ImageDraw.Draw(patch)
radius = max(10, int(min(patch_w, patch_h) * 0.04))
draw.rounded_rectangle(
[(0, 0), (patch_w - 1, patch_h - 1)],
radius=radius,
fill=(255, 255, 255, 245),
)
img_pil.paste(patch, (patch_left, patch_top), patch)
# 官网logo大小:按 patch 短边的 75% 等比缩放
logo_h = max(int(min(patch_w, patch_h) * 0.75), 100)
logo_w = int(logo_h * logo.size[0] / logo.size[1])
logo_r = logo.resize((logo_w, logo_h), Image.LANCZOS)
# Logo 居中放置在覆盖区域
logo_x = patch_left + (patch_w - logo_w) // 2
logo_y = patch_top + (patch_h - logo_h) // 2
img_pil.paste(logo_r, (logo_x, logo_y), logo_r)
img_pil = img_pil.convert("RGB")
img_pil.save(image_path, quality=95)
print(
f" 🏷️ 官网logo叠加完成 (覆盖{scan_w}x{scan_h}→{patch_w}x{patch_h}, "
f"logo:{logo_w}x{logo_h})",
file=sys.stderr,
)
except Exception as e:
logger.error(f"去水印失败 ({image_path}): {e}")
print(f" ⚠️ 去水印失败: {e}", file=sys.stderr)
def _find_image_spots(markdown: str, count: int = 2) -> list:
"""在 markdown 中找适合插入图片的位置,返回 [(插入点索引, 上下文), ...]"""
lines = markdown.split("\n")
candidates = []
# 找段落间的空行位置作为候选
for i in range(1, len(lines) - 1):
# 空行 + 下一行有内容 → 段落边界
if lines[i].strip() == "" and i + 1 < len(lines) and lines[i + 1].strip():
# 确保不在标题前插入(标题通常以 ## 开头)
if not lines[i + 1].strip().startswith("#"):
# 取前后各 3 行作为上下文
ctx_start = max(0, i - 3)
ctx_end = min(len(lines), i + 4)
context = "\n".join(lines[ctx_start:ctx_end])
candidates.append((i + 1, context)) # 在空行后插入
# 均匀选取 count 个点
if not candidates:
return []
if len(candidates) <= count:
return candidates
# 均匀分布:从中间选取
step = len(candidates) // count
selected = [candidates[step * j] for j in range(count)]
return selected
def _insert_images(markdown: str, api_key: str) -> tuple[str, list[str]]:
"""分析 markdown、生成配图并插入,返回 (新 markdown, 生成的图片 web_path 列表)"""
if not IMAGE_GEN_ENABLED:
return markdown, []
spots = _find_image_spots(markdown, IMAGE_GEN_COUNT)
if not spots:
print(" ℹ️ 未找到合适的图片插入位置", file=sys.stderr)
return markdown, []
print(f" 🖼️ 计划插入 {len(spots)} 张配图", file=sys.stderr)
lines = markdown.split("\n")
generated: list[tuple[int, str]] = [] # (line_index, image_path)
for idx, (line_idx, context) in enumerate(spots):
# 生成图片 prompt
img_prompt = _make_image_prompt(context, api_key)
# 输出路径:storage/admin/images/YYYYMM/
month_dir = time.strftime("%Y%m")
filename = f"ai_gen_{int(time.time())}_{idx}.jpg"
output_dir = os.path.join(IMAGE_STORAGE_DIR, month_dir)
output_path = os.path.join(output_dir, filename)
ok = _generate_image(img_prompt, output_path)
if ok:
# logo_add=0 已消除水印,无需后处理覆盖
# _remove_watermark(output_path)
# 相对路径用于 markdown
web_path = f"/storage/admin/images/{month_dir}/{filename}"
generated.append((line_idx, web_path))
else:
print(f" ❌ 图片 {idx+1} 生成失败,跳过", file=sys.stderr)
if not generated:
return markdown, []
# 从后往前插入(避免索引偏移)
for line_idx, web_path in sorted(generated, key=lambda x: x[0], reverse=True):
img_md = f"\n![配图]({web_path})\n"
lines.insert(line_idx, img_md)
# 提取生成的图片路径列表(按插入顺序)
generated_paths = [web_path for _, web_path in sorted(generated, key=lambda x: x[0])]
return "\n".join(lines), generated_paths
# ---------- 微信图片上传 ----------
def _upload_html_images(html: str, publisher) -> str:
"""找出 HTML 中以 /storage/ 开头的本地图片,上传到微信并替换为 CDN 地址"""
import re
from html import unescape
# 匹配 <img src="..." ...> 标签,提取 src 和整个标签
img_pattern = re.compile(r'<img[^>]+src=["\']([^"\']+)["\'][^>]*>', re.IGNORECASE)
storage_base = PROJECT_ROOT
def upload_and_replace(match):
full_tag = match.group(0)
src = match.group(1)
# 只处理以 /storage/ 开头的本地路径
if not src.startswith("/storage/"):
return full_tag
local_path = os.path.normpath(os.path.join(storage_base, src.lstrip("/")))
if not os.path.exists(local_path):
print(f" ⚠️ 图片不存在: {local_path}", file=sys.stderr)
return full_tag
try:
resp = publisher.upload_image(local_path, upload_type="body")
wechat_url = resp.get("url", "")
if wechat_url:
print(f" ✅ 已上传到微信: {os.path.basename(src)}", file=sys.stderr)
# 替换整个 img 标签,只改 src
return full_tag.replace(src, wechat_url)
else:
print(f" ❌ 微信未返回URL: {os.path.basename(src)}", file=sys.stderr)
return full_tag
except Exception as e:
print(f" ❌ 上传失败 {os.path.basename(src)}: {e}", file=sys.stderr)
return full_tag
result = img_pattern.sub(upload_and_replace, html)
return result
# ---------- 核心流程 ----------
def run_pipeline(
article_id: int,
rewritten_content: Optional[str] = None,
) -> Dict[str, Any]:
"""
一键改写→发布流程。
Args:
article_id: CMS 文章 ID
rewritten_content: 预改写好的 markdown。None 则自动调 LLM
Returns:
{"article_id": ..., "title": ..., "media_id": ..., "db_updated": bool}
"""
# 1. 查文章
article = get_article_full(article_id)
if not article:
raise ValueError(f"文章 ID={article_id} 不存在")
title = article["title"]
author = "尿大夫" # 改写后统一使用尿大夫署名
digest = article.get("summary") or ""
original = article["content"]
print(f"📄 文章: ID={article_id} | {title}", file=sys.stderr)
print(f" 原文字数: {len(original)} chars", file=sys.stderr)
# 2. 改写(两种模式)
api_key = None # will be set in auto mode
if rewritten_content:
print(" 模式: 手动改写(跳过 LLM)", file=sys.stderr)
new_content = rewritten_content
else:
# 自动改写前先检查原文内容质量
text_len = _check_original_content(original)
print(f" 原文有效文字: {text_len} chars", file=sys.stderr)
print(" 模式: 自动改写(DeepSeek)", file=sys.stderr)
api_key = load_deepseek_key()
new_content = call_deepseek(original, api_key)
# 检查 LLM 返回是否有效
_check_llm_response(new_content)
print(f" 改写后字数: {len(new_content)} chars", file=sys.stderr)
# 2.1 提取改写后的标题(# 标题)+ 去emoji
rewritten_title, new_content = _extract_title(new_content)
if rewritten_title:
print(f" ✏️ 改写标题: {rewritten_title}", file=sys.stderr)
title = _strip_emoji(rewritten_title)
if title != rewritten_title:
print(f" 🧹 去emoji后: {title}", file=sys.stderr)
else:
print(f" ℹ️ 未检测到改写标题,保留原标题", file=sys.stderr)
# 兜底:原标题也去emoji
title = _strip_emoji(title)
# 2.2 清理第三方链接/作者署名
new_content = _clean_content(new_content)
print(f" 清理后字数: {len(new_content)} chars", file=sys.stderr)
# 2.3 生成配图并插入
print("🎨 生成配图...", file=sys.stderr)
if rewritten_content:
new_content, gen_image_paths = _insert_images(new_content, load_deepseek_key())
else:
new_content, gen_image_paths = _insert_images(new_content, api_key)
# 2.4 末尾添加广告图
new_content = _append_ads_image(new_content)
# 3. 转 HTML
publisher = WeChatPublisher()
html_content = publisher.markdown_to_html(new_content)
# 3.5 上传 HTML 中的本地图片到微信(将 /storage/... 转为微信 CDN 地址)
html_content = _upload_html_images(html_content, publisher)
# 4. 封面图:优先用 AI 生成的配图(已去水印+叠加logo),再降级到 banners/原封面
cover_path = None
cover_web_path = None
if gen_image_paths:
# 解析第一张AI生成图的本地路径
first_web = gen_image_paths[0]
cover_path = os.path.normpath(os.path.join(PROJECT_ROOT, first_web.lstrip("/")))
cover_web_path = first_web
print(f" 🎨 AI配图作为封面: {first_web}", file=sys.stderr)
if not cover_path or not os.path.exists(cover_path):
cover_web_path = None # 本地不存在,不更新DB
cover_path = _random_banner()
if not cover_path:
print(" ℹ️ banners 目录无图,尝试用文章封面", file=sys.stderr)
cover_url = article.get("cover_image") or ""
cover_path = resolve_cover_path(cover_url)
# 5. 创建公众号草稿
print("📝 创建公众号草稿...", file=sys.stderr)
media_id = publisher.create_draft(
title=title,
content=html_content,
author=author,
digest=digest,
cover_image_path=cover_path,
)
print(f" media_id: {media_id}", file=sys.stderr)
# 5.5 根据配置决定是否自动发布
publish_mode = "publish" # 默认值
if os.path.exists(AUTO_CONFIG_PATH):
try:
with open(AUTO_CONFIG_PATH, 'r', encoding='utf-8') as f:
auto_cfg = yaml.safe_load(f)
publish_mode = auto_cfg.get("publish_mode", "publish")
except Exception:
pass
if publish_mode == "publish":
print("📢 自动发布草稿...", file=sys.stderr)
try:
publisher.publish_draft(media_id)
print(" ✅ 已提交发布", file=sys.stderr)
except Exception as e:
print(f" ⚠️ 发布失败: {e}", file=sys.stderr)
else:
print(" 📋 草稿模式: 跳过自动发布,请手动确认", file=sys.stderr)
# 6. 更新数据库(含标题、作者署名、封面图)
ok = update_article_content(
article_id, new_content,
status="published",
new_title=title,
new_author=author,
new_cover_image=cover_web_path,
)
print(f"{'✅' if ok else '❌'} 数据库更新: {'成功' if ok else '失败'}", file=sys.stderr)
return {
"article_id": article_id,
"title": title,
"media_id": media_id,
"db_updated": ok,
}
# ---------- CLI ----------
def main():
parser = argparse.ArgumentParser(
description="CMS 草稿改写 → 公众号发布 一键流水线"
)
parser.add_argument("article_id", type=int, help="CMS 文章 ID")
parser.add_argument(
"--content-file", "-f",
help="改写好的 markdown 文件路径。不传则自动调 DeepSeek 改写"
)
parser.add_argument(
"--dry-run", action="store_true",
help="只改写不发布,输出改写结果到 stdout"
)
args = parser.parse_args()
# 读取改写内容
rewritten = None
if args.content_file:
with open(args.content_file) as f:
rewritten = f.read()
# 干跑模式:只改写输出
if args.dry_run:
article = get_article_full(args.article_id)
if not article:
print(f"❌ 文章 ID={args.article_id} 不存在", file=sys.stderr)
sys.exit(1)
api_key = load_deepseek_key()
result = call_deepseek(article["content"], api_key)
print(result)
return
# 正式执行
try:
result = run_pipeline(args.article_id, rewritten)
print(json.dumps(result, ensure_ascii=False, indent=2), flush=True)
except Exception as e:
print(f"❌ 流水线失败: {e}", file=sys.stderr)
sys.exit(1)
if __name__ == "__main__":
main()
@@ -0,0 +1,515 @@
#!/usr/bin/env python3
"""
CMS 改写 → 公众号发布 一键流水线
两种模式:
1. 自动改写模式 (默认): python pipeline.py <article_id>
自己调 DeepSeek LLM 改写,然后发布
2. 传入改写内容模式: python pipeline.py <article_id> --content-file <markdown文件>
用预改写好的内容直接发布(志玲模式)
环境:
依赖 openai 库: pip install openai
API Key 从 ~/.nanobot/config.json 读取 DeepSeek 配置
"""
import sys
import os
import json
import time
import argparse
from typing import Optional, Dict, Any
# 确保脚本在 rewrite/ 目录执行也能正常工作
SCRIPT_DIR = os.path.dirname(os.path.abspath(__file__))
PROJECT_ROOT = os.path.dirname(SCRIPT_DIR)
BAOYU_DIR = "/home/ubuntu/.nanobot/workspace/skills/baoyu-post-to-wechat"
sys.path.insert(0, BAOYU_DIR)
sys.path.insert(0, SCRIPT_DIR)
from cms_ops import get_article_full, update_article_content
from baoyu_optimized import WeChatPublisher
# ---------- 配置 ----------
MODEL = "deepseek-v4-flash"
NANOBOT_CONFIG = os.path.expanduser("~/.nanobot/config.json")
SYSTEM_PROMPT = """你是一个专业的医疗健康内容编辑。请将下面的文章改写为适合微信公众号发布的版本。
改写要求:
1. **保持医疗准确性**: 不得改变任何医学事实、数据、诊断标准、用药建议
2. **风格口语化**: 用亲近自然的语气,像医生朋友在聊天。适当使用emoji(每段不超过1个)
3. **保留结构**: 保留原文的章节标题和信息层次,但标题可以重新提炼得更吸引人
4. **精简优化**: 去除重复表述,合并相近内容,让文章更紧凑
5. **适合手机阅读**: 段落短小(2-3句一段),善用小标题和列表
6. **只输出改写后的 markdown**,不要加任何解释说明
输出格式: 纯 markdown 文本,可以包含:
- **粗体** 强调关键信息
- 无序列表 -
- ## 二级标题(如果需要)
- 段间空行分隔"""
# ---------- 图片生成配置 ----------
IMAGE_GEN_ENABLED = True # 是否启用配图生成
IMAGE_GEN_COUNT = 2 # 目标生成图片数量(1-3)
IMAGE_STORAGE_DIR = os.path.join(PROJECT_ROOT, "storage", "admin", "images")
IMAGE_PROMPT_SYSTEM = """你是一个图片提示词专家。根据下面给出的文章片段,生成一个适合配图的简短中文描述(用于AI生图),要求:
1. 描述要简洁具体,20-40个字
2. 风格要求:温暖、专业、治愈的视觉风格
3. 内容要与上下文紧密相关
4. 只输出描述文字,不要加任何解释或前缀"""
# 混元生图
HUNYUAN_SECRET_ID = "AKIDlmaJo4Qeq23xmdR7jIn59yd4RZUYUHtP"
HUNYUAN_SECRET_KEY = "QnCSUP6d3Untw6B8tUt4xqkTCoEh1A9u"
HUNYUAN_REGION = "ap-guangzhou"
# Seedream(降级)
SEEDREAM_API_KEY = "83a29955-c633-449f-97f1-f49b62295cf6"
SEEDREAM_BASE_URL = "https://ark.cn-beijing.volces.com/api/v3"
SEEDREAM_MODEL = "doubao-seedream-4-5-251128"
# ---------- 工具函数 ----------
def load_deepseek_key() -> str:
"""从 nanobot 配置读取 DeepSeek API Key"""
try:
with open(NANOBOT_CONFIG) as f:
config = json.load(f)
key = config["providers"]["deepseek"]["apiKey"]
if not key:
raise ValueError("DeepSeek API key 为空")
return key
except Exception as e:
print(f"❌ 读取 DeepSeek API Key 失败: {e}", file=sys.stderr)
print(f" 配置文件: {NANOBOT_CONFIG}", file=sys.stderr)
sys.exit(1)
def call_deepseek(markdown: str, api_key: str) -> str:
"""调用 DeepSeek API 改写文章"""
try:
from openai import OpenAI
except ImportError:
print("❌ 需要安装 openai: pip install openai", file=sys.stderr)
sys.exit(1)
client = OpenAI(api_key=api_key, base_url="https://api.deepseek.com")
print("⏳ 调 DeepSeek 改写中...", file=sys.stderr)
response = client.chat.completions.create(
model=MODEL,
messages=[
{"role": "system", "content": SYSTEM_PROMPT},
{"role": "user", "content": markdown},
],
temperature=0.7,
max_tokens=4096,
)
result = response.choices[0].message.content.strip()
tokens = response.usage.total_tokens if response.usage else "?"
print(f" 消耗 {tokens} tokens", file=sys.stderr)
return result
def resolve_cover_path(cover_url: Optional[str]) -> Optional[str]:
"""将 CMS 封面 URL 转为本地绝对路径"""
if not cover_url:
return None
# cover_url 格式: /storage/admin/images/202605/xxx.jpg
path = os.path.join(PROJECT_ROOT, "..", cover_url.lstrip("/"))
# 某些情况下 cover_url 可能不带前导 /
alt_path = os.path.join(PROJECT_ROOT, "storage", "admin", "images",
*cover_url.split("/")[-2:])
for p in [path, alt_path]:
if os.path.exists(p):
return os.path.normpath(p)
print(f"⚠️ 封面图未找到: {cover_url}", file=sys.stderr)
return None
# ---------- 图片生成 ----------
def _make_image_prompt(context_text: str, api_key: str) -> str:
"""用 LLM 从文章片段生成图片 prompt"""
try:
from openai import OpenAI
except ImportError:
print("⚠️ openai 未安装,使用默认图片 prompt", file=sys.stderr)
return "温暖的医疗健康主题插图,简洁专业的扁平设计风格"
client = OpenAI(api_key=api_key, base_url="https://api.deepseek.com")
try:
response = client.chat.completions.create(
model=MODEL,
messages=[
{"role": "system", "content": IMAGE_PROMPT_SYSTEM},
{"role": "user", "content": context_text},
],
temperature=0.3,
max_tokens=200,
)
prompt = response.choices[0].message.content.strip()
print(f" 🎨 图片 prompt: {prompt}", file=sys.stderr)
return prompt
except Exception as e:
print(f" ⚠️ 图片 prompt 生成失败: {e}", file=sys.stderr)
return "温暖的医疗健康主题插图,简洁专业的扁平设计风格"
def _generate_image(prompt: str, output_path: str,
resolution: str = "1024:1024") -> bool:
"""调用混元生图生成图片,失败则降级到 Seedream"""
success = _hunyuan_image(prompt, output_path, resolution)
if success:
return True
print(" ↪ 降级到 Seedream...", file=sys.stderr)
return _seedream_image(prompt, output_path, resolution)
def _hunyuan_image(prompt: str, output_path: str,
resolution: str = "1024:1024") -> bool:
"""腾讯混元生图 TextToImageLite"""
try:
from tencentcloud.common import credential
from tencentcloud.common.exception.tencent_cloud_sdk_exception import TencentCloudSDKException
from tencentcloud.aiart.v20221229 import aiart_client, models
except ImportError:
return False
try:
cred = credential.Credential(HUNYUAN_SECRET_ID, HUNYUAN_SECRET_KEY)
client = aiart_client.AiartClient(cred, HUNYUAN_REGION)
req = models.TextToImageLiteRequest()
req.Prompt = prompt
req.Resolution = resolution
req.LogoAdd = 0
req.RspImgType = "url"
resp = client.TextToImageLite(req)
result_url = resp.ResultImage
if not result_url:
return False
import requests
r = requests.get(result_url, timeout=60)
r.raise_for_status()
os.makedirs(os.path.dirname(output_path) or ".", exist_ok=True)
with open(output_path, "wb") as f:
f.write(r.content)
print(f" ✅ 混元生图: {os.path.basename(output_path)} ({len(r.content)} bytes)", file=sys.stderr)
return True
except TencentCloudSDKException as e:
print(f" ⚠️ 混元: {e}", file=sys.stderr)
return False
except Exception as e:
print(f" ⚠️ 混元异常: {e}", file=sys.stderr)
return False
def _seedream_image(prompt: str, output_path: str,
resolution: str = "1024:1024") -> bool:
"""Seedream (火山引擎 ARK)"""
import requests
# Seedream 要求 ≥ 3,686,400 像素,自动放大
parts = resolution.replace(":", "x").split("x")
if len(parts) == 2:
try:
w, h = int(parts[0]), int(parts[1])
if w * h < 3686400:
scale = (3686400 / (w * h)) ** 0.5
w2, h2 = (int(w * scale) // 16) * 16, (int(h * scale) // 16) * 16
resolution = f"{w2}x{h2}"
except ValueError:
pass
try:
url = f"{SEEDREAM_BASE_URL}/images/generations"
resp = requests.post(
url,
headers={
"Authorization": f"Bearer {SEEDREAM_API_KEY}",
"Content-Type": "application/json",
},
json={
"model": SEEDREAM_MODEL,
"prompt": prompt,
"size": resolution,
"n": 1,
},
timeout=120,
)
resp.raise_for_status()
data = resp.json()
image_url = data["data"][0].get("url")
if not image_url:
return False
r = requests.get(image_url, timeout=60)
r.raise_for_status()
os.makedirs(os.path.dirname(output_path) or ".", exist_ok=True)
with open(output_path, "wb") as f:
f.write(r.content)
print(f" ✅ Seedream: {os.path.basename(output_path)} ({len(r.content)} bytes)", file=sys.stderr)
return True
except Exception as e:
print(f" ⚠️ Seedream: {e}", file=sys.stderr)
return False
def _find_image_spots(markdown: str, count: int = 2) -> list:
"""在 markdown 中找适合插入图片的位置,返回 [(插入点索引, 上下文), ...]"""
lines = markdown.split("\n")
candidates = []
# 找段落间的空行位置作为候选
for i in range(1, len(lines) - 1):
# 空行 + 下一行有内容 → 段落边界
if lines[i].strip() == "" and i + 1 < len(lines) and lines[i + 1].strip():
# 确保不在标题前插入(标题通常以 ## 开头)
if not lines[i + 1].strip().startswith("#"):
# 取前后各 3 行作为上下文
ctx_start = max(0, i - 3)
ctx_end = min(len(lines), i + 4)
context = "\n".join(lines[ctx_start:ctx_end])
candidates.append((i + 1, context)) # 在空行后插入
# 均匀选取 count 个点
if not candidates:
return []
if len(candidates) <= count:
return candidates
# 均匀分布:从中间选取
step = len(candidates) // count
selected = [candidates[step * j] for j in range(count)]
return selected
def _insert_images(markdown: str, api_key: str) -> str:
"""分析 markdown、生成配图并插入,返回新 markdown"""
if not IMAGE_GEN_ENABLED:
return markdown
spots = _find_image_spots(markdown, IMAGE_GEN_COUNT)
if not spots:
print(" ℹ️ 未找到合适的图片插入位置", file=sys.stderr)
return markdown
print(f" 🖼️ 计划插入 {len(spots)} 张配图", file=sys.stderr)
lines = markdown.split("\n")
generated: list[tuple[int, str]] = [] # (line_index, image_path)
for idx, (line_idx, context) in enumerate(spots):
# 生成图片 prompt
img_prompt = _make_image_prompt(context, api_key)
# 输出路径:storage/admin/images/YYYYMM/
month_dir = time.strftime("%Y%m")
filename = f"ai_gen_{int(time.time())}_{idx}.jpg"
output_dir = os.path.join(IMAGE_STORAGE_DIR, month_dir)
output_path = os.path.join(output_dir, filename)
ok = _generate_image(img_prompt, output_path)
if ok:
# 相对路径用于 markdown
web_path = f"/storage/admin/images/{month_dir}/{filename}"
generated.append((line_idx, web_path))
else:
print(f" ❌ 图片 {idx+1} 生成失败,跳过", file=sys.stderr)
if not generated:
return markdown
# 从后往前插入(避免索引偏移)
for line_idx, web_path in sorted(generated, key=lambda x: x[0], reverse=True):
img_md = f"\n![配图]({web_path})\n"
lines.insert(line_idx, img_md)
return "\n".join(lines)
# ---------- 微信图片上传 ----------
def _upload_html_images(html: str, publisher) -> str:
"""找出 HTML 中以 /storage/ 开头的本地图片,上传到微信并替换为 CDN 地址"""
import re
from html import unescape
# 匹配 <img src="..." ...> 标签,提取 src 和整个标签
img_pattern = re.compile(r'<img[^>]+src=["\']([^"\']+)["\'][^>]*>', re.IGNORECASE)
storage_base = PROJECT_ROOT
def upload_and_replace(match):
full_tag = match.group(0)
src = match.group(1)
# 只处理以 /storage/ 开头的本地路径
if not src.startswith("/storage/"):
return full_tag
local_path = os.path.normpath(os.path.join(storage_base, src.lstrip("/")))
if not os.path.exists(local_path):
print(f" ⚠️ 图片不存在: {local_path}", file=sys.stderr)
return full_tag
try:
resp = publisher.upload_image(local_path, upload_type="body")
wechat_url = resp.get("url", "")
if wechat_url:
print(f" ✅ 已上传到微信: {os.path.basename(src)}", file=sys.stderr)
# 替换整个 img 标签,只改 src
return full_tag.replace(src, wechat_url)
else:
print(f" ❌ 微信未返回URL: {os.path.basename(src)}", file=sys.stderr)
return full_tag
except Exception as e:
print(f" ❌ 上传失败 {os.path.basename(src)}: {e}", file=sys.stderr)
return full_tag
result = img_pattern.sub(upload_and_replace, html)
return result
# ---------- 核心流程 ----------
def run_pipeline(
article_id: int,
rewritten_content: Optional[str] = None,
) -> Dict[str, Any]:
"""
一键改写→发布流程。
Args:
article_id: CMS 文章 ID
rewritten_content: 预改写好的 markdown。None 则自动调 LLM
Returns:
{"article_id": ..., "title": ..., "media_id": ..., "db_updated": bool}
"""
# 1. 查文章
article = get_article_full(article_id)
if not article:
raise ValueError(f"文章 ID={article_id} 不存在")
title = article["title"]
author = article.get("author") or "尿大夫"
digest = article.get("summary") or ""
cover_url = article.get("cover_image") or ""
original = article["content"]
print(f"📄 文章: ID={article_id} | {title}", file=sys.stderr)
print(f" 原文字数: {len(original)} chars", file=sys.stderr)
# 2. 改写(两种模式)
if rewritten_content:
print(" 模式: 手动改写(跳过 LLM)", file=sys.stderr)
new_content = rewritten_content
else:
print(" 模式: 自动改写(DeepSeek)", file=sys.stderr)
api_key = load_deepseek_key()
new_content = call_deepseek(original, api_key)
print(f" 改写后字数: {len(new_content)} chars", file=sys.stderr)
# 2.5 生成配图并插入
print("🎨 生成配图...", file=sys.stderr)
if rewritten_content:
new_content = _insert_images(new_content, load_deepseek_key())
else:
new_content = _insert_images(new_content, api_key)
# 3. 转 HTML
publisher = WeChatPublisher()
html_content = publisher.markdown_to_html(new_content)
# 3.5 上传 HTML 中的本地图片到微信(将 /storage/... 转为微信 CDN 地址)
html_content = _upload_html_images(html_content, publisher)
# 4. 解析封面图路径
cover_path = resolve_cover_path(cover_url)
# 5. 发布公众号草稿
print("🚀 发布公众号草稿...", file=sys.stderr)
media_id = publisher.create_draft(
title=title,
content=html_content,
author=author,
digest=digest,
cover_image_path=cover_path,
)
print(f" media_id: {media_id}", file=sys.stderr)
# 6. 更新数据库
ok = update_article_content(article_id, new_content, status="published")
print(f"{'✅' if ok else '❌'} 数据库更新: {'成功' if ok else '失败'}", file=sys.stderr)
return {
"article_id": article_id,
"title": title,
"media_id": media_id,
"db_updated": ok,
}
# ---------- CLI ----------
def main():
parser = argparse.ArgumentParser(
description="CMS 草稿改写 → 公众号发布 一键流水线"
)
parser.add_argument("article_id", type=int, help="CMS 文章 ID")
parser.add_argument(
"--content-file", "-f",
help="改写好的 markdown 文件路径。不传则自动调 DeepSeek 改写"
)
parser.add_argument(
"--dry-run", action="store_true",
help="只改写不发布,输出改写结果到 stdout"
)
args = parser.parse_args()
# 读取改写内容
rewritten = None
if args.content_file:
with open(args.content_file) as f:
rewritten = f.read()
# 干跑模式:只改写输出
if args.dry_run:
article = get_article_full(args.article_id)
if not article:
print(f"❌ 文章 ID={args.article_id} 不存在", file=sys.stderr)
sys.exit(1)
api_key = load_deepseek_key()
result = call_deepseek(article["content"], api_key)
print(result)
return
# 正式执行
try:
result = run_pipeline(args.article_id, rewritten)
print(json.dumps(result, ensure_ascii=False, indent=2), flush=True)
except Exception as e:
print(f"❌ 流水线失败: {e}", file=sys.stderr)
sys.exit(1)
if __name__ == "__main__":
main()