admin/nanocode
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Auto backup \2026-\04-\20

Browse Source
This commit is contained in:
Ubuntu
2026-07-22 16:00:02 +08:00
parent f7a4a852c2
commit 110e9525ea
4 changed files with 196 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files
new-media-platform/website-new/rewrite/auto_state.json
+4 -10
View File
@@ -1,12 +1,6 @@
{
"date": "2026-06-22",
"fulfilled_slots": [
"08:00"
],
"published_ids": [
4011
],
"used_columns": [
"糖友"
]
"date": "2026-07-01",
"fulfilled_slots": [],
"published_ids": [],
"used_columns": []
}
new-media-platform/website-new/rewrite/cms_ops.py
+68
View File
@@ -198,6 +198,74 @@ def update_article_content(
return affected > 0
def _cleanup_cover_file(cover_image: str):
"""
删除封面图文件及对应缩略图（与后台 file_utils.cleanup_file_by_path 逻辑对齐）。
Args:
cover_image: 封面图路径（/storage/admin/images/... 或相对路径）
"""
if not cover_image or cover_image.startswith(("http://", "https://")):
return
upload_dir = os.path.join(PROJECT_ROOT, "storage", "admin", "images")
thumbnail_dir = os.path.join(upload_dir, "thumbnails")
# 兼容两种路径格式
if cover_image.startswith("/storage/admin/images/"):
rel_path = cover_image[len("/storage/admin/images/"):]
else:
rel_path = cover_image.lstrip("/")
full_path = os.path.join(upload_dir, rel_path)
# 删除原文件
try:
if os.path.isfile(full_path):
os.remove(full_path)
except OSError:
pass
# 同步删除缩略图（与原图同名，扩展名 .jpg）
try:
stem = os.path.splitext(os.path.basename(rel_path))[0]
thumb_path = os.path.join(thumbnail_dir, f"{stem}.jpg")
if os.path.isfile(thumb_path):
os.remove(thumb_path)
except OSError:
pass
def delete_article(article_id: int) -> bool:
"""
删除文章（物理删除），同步清理封面图片文件及缩略图。
与后台 routers/articles.py 的 delete_article 行为对齐。
Args:
article_id: 文章 ID
Returns:
是否删除成功
"""
conn = _get_conn()
cursor = conn.cursor()
# 先查 cover_image 以便删除后清理文件
cursor.execute("SELECT cover_image FROM articles WHERE id = ?", (article_id,))
row = cursor.fetchone()
cover_image = row["cover_image"] if row else None
cursor.execute("DELETE FROM articles WHERE id = ?", (article_id,))
conn.commit()
affected = cursor.rowcount
conn.close()
if affected > 0 and cover_image:
_cleanup_cover_file(cover_image)
return affected > 0
def count_drafts() -> int:
"""统计草稿文章数量"""
conn = _get_conn()
new-media-platform/website-new/rewrite/pipeline.py
+79 -10
View File
@@ -33,7 +33,7 @@ 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 cms_ops import get_article_full, update_article_content, delete_article
from baoyu_optimized import WeChatPublisher
# ---------- 配置 ----------
@@ -76,6 +76,8 @@ IMAGE_PROMPT_SYSTEM = """你是一个图片提示词专家。根据下面给出
HUNYUAN_API_KEY = "sk-qgPNJwuOGji7I7ixqNbbkWaTIvF1MrYTRWQnEZ3ZRVU4u7y3"
HUNYUAN_API_BASE = "https://tokenhub.tencentmaas.com/v1/api/image"
HUNYUAN_MODEL = "hy-image-v3.0"
HUNYUAN_LITE_MODEL = "hy-image-lite"
HUNYUAN_LITE_ENDPOINT = "https://tokenhub.tencentmaas.com/v1/api/image/lite"
# Seedream（降级）
SEEDREAM_API_KEY = "83a29955-c633-449f-97f1-f49b62295cf6"
@@ -297,10 +299,10 @@ def _strip_images(markdown: str) -> str:
return re.sub(r'!\[.*?\]\([^)]+\)', '', markdown)
def _check_original_content(content: str) -> int:
def _check_original_content(content: str, article_id: int) -> 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。
返回纯文本字符数；低于 _MIN_TEXT_LENGTH 删除文章并抛出 ValueError。
"""
text = _strip_images(content)
# 也移除 markdown 链接、HTML 标签等
@@ -309,27 +311,33 @@ def _check_original_content(content: str) -> int:
text = re.sub(r'\s+', ' ', text).strip()
text_len = len(text)
if text_len < _MIN_TEXT_LENGTH:
delete_article(article_id)
print(f" 🗑️ 已删除短文章 ID={article_id}（{text_len} 字符）", file=sys.stderr)
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"
f"可能全部为图片，已删除"
)
return text_len
def _check_llm_response(content: str) -> None:
def _check_llm_response(content: str, article_id: int) -> 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):
delete_article(article_id)
print(f" 🗑️ 已删除占位回复文章 ID={article_id}", file=sys.stderr)
raise ValueError(
f"LLM 返回疑似占位回复（匹配模式: {pattern}），不发布"
f"LLM 返回疑似占位回复（匹配模式: {pattern}），已删除"
)
# 额外检查：改写后内容过短（< 200 字）也视为无效
if len(content) < _MIN_TEXT_LENGTH:
delete_article(article_id)
print(f" 🗑️ 已删除改写后过短文章 ID={article_id}（{len(content)} 字符）", file=sys.stderr)
raise ValueError(
f"改写后内容过短（{len(content)} 字符 < {_MIN_TEXT_LENGTH}），不发布"
f"改写后内容过短（{len(content)} 字符 < {_MIN_TEXT_LENGTH}），已删除"
)
@@ -374,12 +382,73 @@ def _make_image_prompt(context_text: str, api_key: str) -> str:
return "温暖的医疗健康主题插图，简洁专业的扁平设计风格"
def _hunyuan_lite_image(prompt: str, output_path: str,
resolution: str = "1024:1024") -> bool:
"""腾讯混元生图 hy-image-lite (轻量同步版)
POST 一次直接返回图片 URL，无需轮询。价格更低。
注意：Lite 固定输出 1024×1024，不支持自定义分辨率。
"""
import requests
if not HUNYUAN_API_KEY:
return False
try:
payload = {
"model": HUNYUAN_LITE_MODEL,
"prompt": prompt,
"rsp_img_type": "url",
}
headers = {
"Authorization": f"Bearer {HUNYUAN_API_KEY}",
"Content-Type": "application/json",
}
resp = requests.post(HUNYUAN_LITE_ENDPOINT, json=payload,
headers=headers, timeout=60)
resp.raise_for_status()
result = resp.json()
data = result.get("data", [])
if not data:
print(f" ⚠️ 混元 Lite 无数据: {result}", file=sys.stderr)
return False
img_url = data[0].get("url", "")
if not img_url:
print(f" ⚠️ 混元 Lite 无 URL: {data[0]}", file=sys.stderr)
return False
r = requests.get(img_url, timeout=60)
r.raise_for_status()
img_data = r.content
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 Lite: {os.path.basename(output_path)} ({len(img_data)} bytes)", file=sys.stderr)
return True
except Exception as e:
print(f" ⚠️ 混元 Lite 异常: {e}", file=sys.stderr)
return False
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"""
"""生成封面配图：优先混元 Lite → 混元 V3.0 → Seedream"""
# 1) 混元 Lite (轻量版，更便宜，同步返回)
success = _hunyuan_lite_image(prompt, output_path, resolution)
if success:
return True
# 2) 混元 V3.0 (标准版)
print(" ↪ 降级到混元 V3.0...", file=sys.stderr)
success = _hunyuan_image(prompt, output_path, resolution)
if success:
return True
# 3) Seedream
print(" ↪ 降级到 Seedream...", file=sys.stderr)
return _seedream_image(prompt, output_path, resolution)
@@ -806,13 +875,13 @@ def run_pipeline(
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)
text_len = _check_original_content(original, article_id)
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)
_check_llm_response(new_content, article_id)
print(f" 改写后字数: {len(new_content)} chars", file=sys.stderr)