Identify Breaking News

Identify articles that qualify as breaking news requiring immediate notification.

When to Use

Use this skill when you need to:

• Filter analyzed articles for breaking news
• Determine which stories warrant immediate alerts
• Prioritize high-impact news for real-time notification
• Avoid alert fatigue from non-critical news

Instructions

Step 1: Search for High-Importance Articles

Use memory/search to find recent high-importance analyses:

• namespace: "news/analyses"
• query: "breaking news important announcement"
• limit: 50

Filter results for:

• importance_score >= 8
• is_breaking == true
• Recent timestamp (last 24 hours)

Step 2: Check Notification History

For each potential breaking article, use memory/check_seen:

• key: article content hash or URL
• namespace: "news/breaking_alerts"

Skip articles that have already been notified.

Step 3: Calculate Breaking Score

For articles not yet notified, calculate breaking score:

Score components (0-100 total):

• Importance (0-50): importance_score * 5
• Source credibility (0-20): +20 for authoritative sources
• LLM breaking flag (0-20): +20 if is_breaking=true
• Category bonus (0-10): +10 for security/product/research
• Keyword bonus (0-10): +2 per breaking keyword in title

Authoritative sources:

• OpenAI Blog, Anthropic Blog, Google AI Blog
• Meta AI Blog, Microsoft AI Blog, DeepMind
• arXiv, Nature, Science

Breaking keywords:

• release, launch, announce, breakthrough
• vulnerability, security, critical, urgent
• breaking, major, significant

Step 4: Apply Threshold

Filter articles by breaking score threshold:

• Score 90-100: Critical breaking news - immediate notification
• Score 80-89: High-priority - notify within 15 minutes
• Score 70-79: Notable - notify within 1 hour
• Score <70: Not breaking - include in regular digest

Step 5: Apply Rate Limiting

Prevent alert fatigue by limiting notifications:

• Maximum 5 per hour
• Maximum 20 per day

If rate limit exceeded, queue for later notification.

Step 6: Mark as Notified

For each article being notified, use memory/mark_seen:

• key: article content hash or URL
• namespace: "news/breaking_alerts"
• ttl_seconds: 86400 (24 hours)

Step 7: Return Results

Return breaking news articles including:

• Article details (title, source, summary, URL)
• Breaking score
• Notification priority
• Breaking reason (if available)

Tool Usage Guidance

memory/search

• Search for high-importance analyses
• Filter by importance_score and is_breaking
• Limit to recent timeframe

memory/check_seen

• Check if article already notified
• Use content hash as key
• Prevents duplicate alerts

memory/mark_seen

• Mark after sending notification
• Use 24-hour TTL
• Allows re-notification after window

Breaking Score Components

Threshold Guidelines

Score 90-100: Critical

• Major model releases (GPT-5, Claude 4)
• Critical security vulnerabilities
• Regulatory decisions with immediate impact
• Action: Immediate notification

Score 80-89: High-Priority

• Significant product launches
• Important research breakthroughs
• Major partnerships or acquisitions
• Action: Notify within 15 minutes

Score 70-79: Notable

• Meaningful product updates
• Interesting research results
• Industry developments
• Action: Notify within 1 hour

Score <70: Not Breaking

• Regular news, incremental updates
• Action: Include in regular digest only

Breaking News Data Schema

{
  "url": "https://openai.com/blog/gpt-5",
  "title": "Introducing GPT-5",
  "source": "OpenAI Blog",
  "ai_summary": "OpenAI has released GPT-5 with major advances...",
  "importance_score": 10,
  "is_breaking": true,
  "breaking_reason": "Major model release from leading AI company",
  "breaking_score": 95,
  "category": "product",
  "content_hash": "abc123def456"
}

Error Handling

• If search returns no results, return empty list
• If check_seen fails, proceed with notification (better duplicate than missed)
• If mark_seen fails, log warning but continue

Success Criteria

• Breaking news identified within 15 minutes of publication
• False positive rate < 10%
• No duplicate alerts for the same story
• Rate limits prevent alert fatigue
• Critical stories (score >90) never missed