ai-news-digest（AI 新闻文摘）

AI News Digest Skill

Overview

This skill automatically aggregates AI news from 8+ authoritative sources (both English and Chinese), intelligently deduplicates articles, ranks them by importance, and delivers formatted digests via email with Markdown attachments.

Key Features:

• 🌐 Multi-source aggregation (RSS-based, no scraping issues)
• 🧹 Smart deduplication (similarity detection)
• 📊 Intelligent ranking (heat score, recency, importance)
• 📧 Email delivery with HTML + Markdown attachment
• 📝 Auto-save to Obsidian vault
• ⏰ Flexible time ranges (hourly, daily, weekly)

Quick Start

Generate and Email Today's Digest

cd /data/workspace/TsunagiConfig/skills/ai-news-digest
python scripts/generate_digest.py

This will:

1. Fetch news from the past 24 hours
2. Deduplicate and rank articles
3. Generate HTML email + Markdown file
4. Send to your configured email
5. Save to Obsidian (if configured)

Get Last 6 Hours of Breaking News

python scripts/generate_digest.py --hours 6

Weekly Summary (Last 7 Days)

python scripts/generate_digest.py --days 7

News Sources

All sources use RSS feeds for stability (no web scraping SSL issues).

Usage Examples

1. Manual One-Time Digest

# Default: last 24 hours, send email, save to Obsidian
python scripts/generate_digest.py

# Last 12 hours, no email
python scripts/generate_digest.py --hours 12 --no-email

# Last 3 days, custom output path
python scripts/generate_digest.py --days 3 --output ~/Desktop/ai_digest.md

2. Scheduled Daily Digest (Cron)

# Add to crontab: Daily at 8 AM
0 8 * * * cd /data/workspace/TsunagiConfig/skills/ai-news-digest && python scripts/generate_digest.py --hours 24

3. Use as Python Module

from scripts.fetch_news import fetch_all_news
from scripts.process_news import deduplicate_and_rank

# Fetch news from last 48 hours
articles = fetch_all_news(hours=48)
print(f"Fetched {len(articles)} articles")

# Process and rank
top_articles = deduplicate_and_rank(articles, top_n=15)

# Generate Markdown
from scripts.generate_digest import format_markdown
markdown = format_markdown(top_articles)
print(markdown)

4. Email Only (No File Output)

from scripts.generate_digest import send_digest_email

articles = [...]  # Your processed articles
send_digest_email(
    articles=articles,
    recipient="your@email.com",
    subject="AI News Digest - 2024-01-15"
)

Configuration

Email Settings (Optional)

Edit config/email.json if you want to add email delivery:

{
  "smtp_host": "smtp.example.com",
  "smtp_port": 587,
  "sender": "your-email@example.com",
  "auth_code": "your-smtp-auth-code",
  "recipient": "recipient@example.com"
}

Note: Email functionality is optional. The skill works without it, generating Markdown reports only.

News Sources

Edit config/sources.json to add/remove sources:

{
  "sources": [
    {
      "name": "Hacker News",
      "url": "https://hn.algolia.com/api/v1/search_by_date?tags=story&query=AI",
      "type": "api",
      "language": "en",
      "enabled": true
    },
    {
      "name": "机器之心",
      "url": "https://www.jiqizhixin.com/rss",
      "type": "rss",
      "language": "zh",
      "enabled": true
    }
  ]
}

Obsidian Integration

Edit config/obsidian.json:

{
  "vault_path": "/data/workspace/obsidian/Daily Notes",
  "enabled": true,
  "filename_format": "AI Digest - {date}.md"
}

Output Format

Markdown Structure

# AI News Digest - 2024-01-15

**Time Range:** Last 24 hours
**Total Articles:** 38 → 15 after deduplication
**Generated:** 2024-01-15 08:00:00

---

## 🔥 Top Stories (Score ≥ 8.0)

### 1. OpenAI Announces GPT-5 with Multimodal Reasoning
- **Source:** OpenAI Blog | **Published:** 2 hours ago
- **Score:** 9.2/10 | **Category:** Product Launch
- **Summary:** OpenAI unveiled GPT-5, featuring advanced multimodal reasoning capabilities across text, images, and audio. The model shows significant improvements in complex problem-solving tasks.
- **Link:** [Read More](https://openai.com/blog/gpt-5-announcement)

---

## 📊 Industry News (Score 6.0-7.9)

### 2. Google DeepMind's AlphaFold 3 Predicts Protein Structures
...

---

## 🌏 Chinese AI Updates

### 8. 字节跳动发布新一代大模型"豆包Pro"
- **来源:** 量子位 | **发布:** 5小时前
...

---

## 📈 Statistics
- Total sources checked: 8
- Articles fetched: 38
- After deduplication: 15
- English articles: 10
- Chinese articles: 5

Email HTML Format

• Clean, responsive HTML design
• Clickable headlines
• Category badges (🔥 Hot, 📊 Industry, 🌏 Chinese)
• Markdown file attached as .md

Command-Line Options

python scripts/generate_digest.py [OPTIONS]

Options:
  --hours N          Time range in hours (default: 24)
  --days N           Time range in days (overrides --hours)
  --output PATH      Output Markdown file path
  --send-email       Send via email (default: true)
  --no-email         Skip email sending
  --save-obsidian    Save to Obsidian (default: true if configured)
  --top-n N          Number of top articles to include (default: 15)
  --min-score FLOAT  Minimum score threshold (default: 5.0)
  --debug            Enable debug logging

Common Tasks

Test News Fetching

python scripts/test_digest.py

Output:

✓ Fetched 16 articles from Hacker News
✓ Fetched 22 articles from 机器之心
✓ Total: 38 articles
✓ After deduplication: 15 articles

Debug RSS Feed Issues

from scripts.fetch_news import test_single_source

# Test a specific source
test_single_source("https://www.jiqizhixin.com/rss")

Custom Scoring Algorithm

Edit scripts/process_news.py:

def calculate_score(article):
    score = 0.0

    # Recency (0-3 points)
    hours_ago = (datetime.now() - article['published']).total_seconds() / 3600
    if hours_ago < 6:
        score += 3.0
    elif hours_ago < 24:
        score += 2.0
    else:
        score += 1.0

    # Keyword boost (0-4 points)
    hot_keywords = ['GPT', 'OpenAI', 'breakthrough', 'AGI']
    for keyword in hot_keywords:
        if keyword.lower() in article['title'].lower():
            score += 1.0

    # Source authority (0-3 points)
    authority_map = {
        'OpenAI Blog': 3.0,
        'Hacker News': 2.5,
        '机器之心': 2.5
    }
    score += authority_map.get(article['source'], 1.0)

    return min(score, 10.0)

Troubleshooting

Issue: No articles fetched

# Check internet connectivity
curl -I https://www.jiqizhixin.com/rss

# Test RSS parsing
python -c "import feedparser; print(feedparser.parse('https://www.jiqizhixin.com/rss'))"

Issue: Email not sending

# Test email config
node /data/workspace/send_report_email.js

# Check SMTP credentials in config/email.json

Issue: Too many duplicate articles

Adjust similarity threshold in scripts/process_news.py:

SIMILARITY_THRESHOLD = 0.75  # Lower = stricter deduplication (default: 0.8)

Quick Reference

Installation

Before using this skill, install the required dependencies:

pip install -r requirements.txt

This will install:

• feedparser - RSS feed parsing
• beautifulsoup4 - HTML content extraction
• requests - HTTP requests
• python-dateutil - Date/time handling
• numpy - Similarity calculations

System Requirements:

• Python 3.8+
• Internet connection
• Optional: Node.js 14+ (for email sending via nodemailer)

Next Steps

• For API reference and data structures, see references/api_reference.md
• For advanced customization, see references/advanced_usage.md
• To add new news sources, edit config/sources.json
• To integrate with other tools, use the Python module API

License

MIT License - Free to use and modify