News Aggregator Skill

Supported Data Sources

hackernews

weibo

github

36kr

producthunt

v2ex

tencent

wallstreetcn

Tool Usage

Basic Commands

uv run --directory .agents/skills/news-aggregator-skill python scripts/fetch_news.py [parameters]

Parameter Description

• --source <source>

  : Specify data source
  • Single source:

    hackernews

    ,

    weibo

    ,

    github

    ,

    36kr

    ,

    producthunt

    ,

    v2ex

    ,

    tencent

    ,

    wallstreetcn

  • Multiple sources (comma separated):

    hackernews,github,producthunt

  • All sources:

    all

• --limit <count>

  : Maximum number of entries returned per source (default: 10)

• --keyword <keyword>

  : Keyword filtering (comma separated)
  • Example:

    "AI,LLM,GPT"

  • Case insensitive, supports word boundary matching

• --deep

  : Enable deep scraping
  • Download and extract article body content (truncate first 3000 characters)
  • Concurrent scraping to improve speed
  • The result will include the

    content

    field

Output Format

• source

  : Source name

• title

  : Title

• url

  : Link

• heat

  : Heat indicator (points, number of replies, star count, etc.)

• time

  : Time information

• content

  : Article content (only available when

  --deep

  is used)

Usage Strategies

1. Global Scan (Wide Acquisition)

# Fetch from all sources, 15 entries per source, enable deep scraping
uv run --directory .agents/skills/news-aggregator-skill python scripts/fetch_news.py --source all --limit 15 --deep

2. Single Data Source

# Top 10 entries from Hacker News
uv run --directory .agents/skills/news-aggregator-skill python scripts/fetch_news.py --source hackernews --limit 10 --deep

# Top 15 entries from GitHub Trending
uv run --directory .agents/skills/news-aggregator-skill python scripts/fetch_news.py --source github --limit 15 --deep

3. Keyword Search (Intelligent Expansion)

• User says "AI" → Use:

  "AI,LLM,GPT,Claude,DeepSeek,Gemini,机器学习,RAG,Agent,大模型"

• User says "frontend" → Use:

  "前端,React,Vue,Next.js,TypeScript,JavaScript,CSS,Vite"

• User says "finance" → Use:

  "金融,股票,市场,经济,加密货币,比特币,黄金,A股"

# Example: User asks "What AI related news are there"
uv run --directory .agents/skills/news-aggregator-skill python scripts/fetch_news.py \
  --source hackernews,github,36kr \
  --limit 20 \
  --keyword "AI,LLM,GPT,Claude,DeepSeek,Agent,大模型" \
  --deep

4. Precise Search

# Search news related to "DeepSeek"
uv run --directory .agents/skills/news-aggregator-skill python scripts/fetch_news.py --source all --limit 10 --keyword "DeepSeek" --deep

Output Specifications

Report Format Requirements

• Use Simplified Chinese
• Adopt magazine/newsletter style (such as The Economist or Morning Brew)
• Professional, concise and engaging

1. Front Page Headlines (3-5 entries)
   • The most important cross-domain news
2. Technology & AI
   • Special section for AI, LLM and technology related content
3. Finance/Society
   • Other important categories (sorted by relevance)

### Number. [Title Text](Original URL)

**Source**: <Data Source> | **Time**: <Time Information> | **Heat**: <Heat Indicator>

**Key Point**: One sentence summary of "so what?"

**In-depth Interpretation**:
- Point 1: Why it is important
- Point 2: Technical details or background
- Point 3: Impact and implications

• ✅ Title must be a Markdown link:

  [OpenAI releases GPT-5](https://...)

• ❌ Plain text titles are prohibited:

  OpenAI releases GPT-5

• Metadata line must include: source, time, heat
• 2-3 interpretation points must be provided when deep scanning is enabled

Time Filtering and Intelligent Supplement

1. Prioritize user's window: First list entries that strictly meet the time requirements
2. Intelligent supplement: If the list is too short, must include high-value/hot entries in a larger range (such as the past 24 hours)
3. Clear marking: Clearly mark supplementary entries (such as "⚠️ 18h ago", "🔥 24h Hot")
4. Value first: Even if it slightly exceeds the time window, prioritize displaying SOTA, major releases or high-heat content

• Strictly return valid entries in the crawl list (such as Top 10)
• List all crawled entries
• Do not perform intelligent supplementation
• Must conduct in-depth analysis for each project:
  • Core Value: What problem does it solve? Why is it popular?
  • Inspirational Thinking: Technical or product insights
  • Scenario Tags: 3-5 keywords (such as

    #RAG #Local First #Rust

    )

Output File

• Save Location:

  reports/

  folder in the workspace root directory
• File Naming: With timestamp (such as

  hn_news_20260131_1430.md

  )
• Full Path Example:

  /Users/dio/Documents/new_vault/reports/tech_news_20260131_1430.md

• User Display: Present the full report content in the chat

Interactive Menu

1. Read the

   templates.md

   file in the skill directory
2. Show the user the list of available commands in the file
3. Guide the user to select the number or copy the command to execute