feat(writer): load previous blog posts and provide Chinese copywriting guidelines

Author: at-wr

.github/workflows/auto-writer.yml

@@ -13,7 +13,7 @@ jobs:
         uses: actions/checkout@v4
       - name: Install Python libraries
         run: |
-          pip install openai bs4 requests
+          pip install openai bs4 requests yaml
       - name: Article creation script
         env:
           DS_APIKEY: ${{ secrets.DS_APIKEY }}

writer.py

@@ -5,65 +5,95 @@
 import time
 import datetime
 import os
+import glob
+import yaml
 
 path_to = f'src/content/blog/{datetime.datetime.now().strftime("%Y-%m-%d")}'
 
 if os.path.exists(path_to):
-	print("Article already generated today.")
-	exit(0)
+    print("Article already generated today.")
+    exit(0)
 else:
-	os.makedirs(path_to, exist_ok=True)
-	print(f"Created directory {path_to}")
+    os.makedirs(path_to, exist_ok=True)
+    print(f"Created directory {path_to}")
 
 start = time.time()
 print("Connecting to LLM API ...")
 deepseek = OpenAI(base_url="https://api.deepseek.com", api_key=os.environ.get("DS_APIKEY"))
 print(f"Initialized LLM API. ({time.time() - start:.1f}s)")
 
 def generate(context, provider, model):
-	completion = provider.chat.completions.create(
-		model=model,
-		messages=context
-	)
-	return completion.choices[0].message.content.strip()
+    completion = provider.chat.completions.create(
+        model=model,
+        messages=context
+    )
+    return completion.choices[0].message.content.strip()
 
 def scrape_website(url, css_selector):
-	response = requests.get(url)
-	if response.status_code == 200:
-		soup = BeautifulSoup(response.content, "html.parser")
-		elements = soup.select(css_selector)
-		return elements
-	else: return []
+    response = requests.get(url)
+    if response.status_code == 200:
+        soup = BeautifulSoup(response.content, "html.parser")
+        elements = soup.select(css_selector)
+        return elements
+    else: return []
+
+# New function to get existing blog posts
+def get_existing_blog_posts():
+    blog_posts = []
+    blog_paths = glob.glob("src/content/blog/*/index.md")
+    
+    for path in blog_paths:
+        try:
+            with open(path, 'r', encoding='utf-8') as f:
+                content = f.read()
+                # Parse frontmatter
+                if content.startswith('---'):
+                    _, frontmatter, markdown = content.split('---', 2)
+                    metadata = yaml.safe_load(frontmatter)
+                    blog_posts.append({
+                        'title': metadata.get('title', ''),
+                        'description': metadata.get('description', '')
+                    })
+        except Exception as e:
+            print(f"Error reading {path}: {e}")
+    
+    return blog_posts
+
+# Get existing blog posts
+existing_posts = get_existing_blog_posts()
+existing_posts_text = "\n".join([f"标题: {post['title']}\n描述: {post['description']}" for post in existing_posts])
+print(f"Loaded {len(existing_posts)} existing blog posts.")
 
 topics = [topic.get_text(strip=True) for topic in scrape_website("https://news.ycombinator.com/", ".titleline")]
 topics_text = "\n".join(random.choices(topics, k=random.randint(5, len(topics))))
 print(f"Scraped {len(topics)} topics from Hacker News.")
 
 def extract_topic(topics):
-	global deepseek
-	return generate([
-		{"role": "system", "content": "你在为一篇技术博客确定一个主题。直接用中文输出主题。"},
-		{"role": "user", "content": f"阅读以下是HackerNews的热门文章，然后写一个可以用于技术博客的主题。这个主题应当是一个通用、普通的技术，不能是一个事件或其它东西。\n\n{topics}\n\n只需要一个主题，直接输出。"},
-	], deepseek, "deepseek-chat")
+    global deepseek, existing_posts_text
+    return generate([
+        {"role": "system", "content": "你在为一篇技术博客确定一个主题。直接用中文输出主题。"},
+        {"role": "user", "content": f"阅读以下是HackerNews的热门文章，然后写一个可以用于技术博客的主题。这个主题应当是一个通用、普通的技术，不能是一个事件或其它东西。\n\n{topics}\n\n以下是已有的博客文章，请避免选择相似的主题：\n\n{existing_posts_text}\n\n只需要一个主题，直接输出。"},
+    ], deepseek, "deepseek-chat")
 
 def outline(topic):
-	global deepseek
-	return generate([
-		{"role": "user", "content": f"我要写一篇关于「{topic}」的博客文章。帮我列一个详细的文章提纲。"}
-	], deepseek, "deepseek-reasoner")
+    global deepseek
+    return generate([
+        {"role": "user", "content": f"我要写一篇关于「{topic}」的博客文章。帮我列一个详细的文章提纲。"}
+    ], deepseek, "deepseek-reasoner")
 
 def write_from_outline(outline):
-	global deepseek
-	return generate([
-		{"role": "user", "content": f"{outline}\n\n根据这个提纲中关于技术知识的部分，写出一篇技术博客文章。文章中避免出现图片，避免使用列表。每一段出现的代码都进行较为详细的解读。在讲述内容时尽量使用段落的语言，语言风格可以略偏专业，但保持清晰。使用Markdown（要求符合Common Markdown规范）输出，使用LaTeX公式（注意：数学的开闭定界符前后不能有字母或数字字符。像x$a + b = c$或$a + b = c$1将无法渲染为数学公式（所有$会被渲染为$）；但x $\infty$ 1和($\infty$)会正常渲染），标题尽量只用一级标题 `#` 和二级标题 `##`，不要用分割线。直接输出正文。"}
-	], deepseek, "deepseek-reasoner")
+    global deepseek, existing_posts_text
+    return generate([
+        {"role": "system", "content": "你是一位专业技术博客作者。在写作时请遵循以下中文排版规范：1) 中文与英文、数字之间需要有空格；2) 中文标点与英文、数字之间不加空格；3) 使用全角中文标点；4) 专有名词大小写正确；5) 英文、数字使用半角字符；6) 使用直角引号「」。"},
+        {"role": "user", "content": f"{outline}\n\n根据这个提纲中关于技术知识的部分，写出一篇技术博客文章。文章中避免出现图片，避免使用列表。每一段出现的代码都进行较为详细的解读。在讲述内容时尽量使用段落的语言，语言风格可以略偏专业，但保持清晰。\n\n以下是已有的博客文章，请确保你的内容与它们不重复：\n\n{existing_posts_text}\n\n使用Markdown（要求符合Common Markdown规范）输出，使用LaTeX公式（注意：数学的开闭定界符前后不能有字母或数字字符。像x$a + b = c$或$a + b = c$1将无法渲染为数学公式（所有$会被渲染为$）；但x $\\infty$ 1和($\\infty$)会正常渲染），标题尽量只用一级标题 `#` 和二级标题 `##`，不要用分割线。直接输出正文。"}
+    ], deepseek, "deepseek-reasoner")
 
 def summary(article):
-	global deepseek
-	return generate([
-		{"role": "system", "content": "你是一个技术博客简介写作者，简介不一定需要涵盖文章的全部内容，能起到一定的提示作用即可。直接输出简介。"},
-		{"role": "user", "content": f"给这篇文章写一个15字的简短介绍：\n\n{article}"}
-	], deepseek, "deepseek-chat")
+    global deepseek
+    return generate([
+        {"role": "system", "content": "你是一个技术博客简介写作者，简介不一定需要涵盖文章的全部内容，能起到一定的提示作用即可。直接输出简介。请遵循中文排版规范，确保中英文之间有空格，使用正确的标点符号。"},
+        {"role": "user", "content": f"给这篇文章写一个15字的简短介绍：\n\n{article}"}
+    ], deepseek, "deepseek-chat")
 
 start = time.time()
 print("Generating topic ...")
@@ -90,30 +120,30 @@ def summary(article):
 author = random.choice(["杨其臻", "杨子凡", "叶家炜", "黄京"])
 
 for line in lines:
-	if line.startswith("# "):
-		title = line[1:].strip()
-		print(f"Detected title: {title}")
-
-		metadata = "\n".join([
-			"---",
-			f'title: "{title}"',
-			f'author: "{author}"',
-			f'date: "{datetime.datetime.now().strftime("%b %d, %Y")}"',
-			f'description: "{summary_result}"',
-			f'latex: true',
-			f'pdf: true',
-			"---",
-		]) + "\n"
-		print(f"Injecting metadata:\n{metadata.strip()}")
-
-		markdown_file += metadata
-		break
+    if line.startswith("# "):
+        title = line[1:].strip()
+        print(f"Detected title: {title}")
+
+        metadata = "\n".join([
+            "---",
+            f'title: "{title}"',
+            f'author: "{author}"',
+            f'date: "{datetime.datetime.now().strftime("%b %d, %Y")}"',
+            f'description: "{summary_result}"',
+            f'latex: true',
+            f'pdf: true',
+            "---",
+        ]) + "\n"
+        print(f"Injecting metadata:\n{metadata.strip()}")
+
+        markdown_file += metadata
+        break
 
 for line in lines:
-	if line.startswith("---"): continue
-	markdown_file += line + "\n"
+    if line.startswith("---"): continue
+    markdown_file += line + "\n"
 
 with open(f"{path_to}/index.md", "w", encoding="utf-8") as f:
-	f.write(markdown_file)
+    f.write(markdown_file)
 
 print(f"Markdown file generated at {path_to}/index.md")