analyze-trends

X-McKay 的头像作者 X-McKay

Analyze patterns and trends across a batch of articles. Identifies hot topics covered by multiple sources, tracks topic momentum, and detects emerging themes.

📖 中文导读:Analyze patterns and trends across a batch of articles. Identifies hot topics covered by multiple sources, tracks topic momentum, and detects emerging themes. 以下正文为该技能的原始 SKILL.md 内容(英文),上方安装命令可直接使用。

name: analyze-trends description: > Analyze patterns and trends across a batch of articles. Identifies hot topics covered by multiple sources, tracks topic momentum, and detects emerging themes. metadata: domain: news category: diagnostic mcp-servers: [] requires-approval: false confidence: 0.8 input:

  • name: articles type: list[ProcessedArticle] description: Processed articles with entities extracted
  • name: hot_threshold type: int default: 3 description: Minimum sources for a topic to be considered "hot"
  • name: lookback_days type: int default: 7 description: Days to look back for historical trend data output:
  • name: trends type: list[TrendingTopic] description: Identified trends with status (hot/rising/established/fading)
  • name: entity_clusters type: list[EntityCluster] description: Groups of entities that frequently appear together

Analyze Trends

Identify trending topics and patterns across news articles.

Preconditions

  • Articles have entities extracted by analyze-article skill
  • Memory system (Qdrant) available for historical comparison

Actions

Step 1: Extract Topics from Entities

Group articles by shared entities:

  • Collect all entities from all articles
  • Filter entities shorter than 3 characters
  • Map each entity to list of articles mentioning it
  • Keep only entities appearing in 2+ articles

Step 2: Detect Hot Topics

For each topic (entity):

  1. Count unique sources covering it
  2. If sources >= hot_threshold (default 3), mark as HOT
  3. Calculate momentum based on source coverage

Step 3: Compare to Historical Data

Query memory for historical theme data from past lookback_days:

  • Calculate trend status based on comparison:
    • BREAKING: New topic with high immediate coverage
    • HOT: Multiple sources covering simultaneously
    • RISING: Increasing coverage compared to history
    • ESTABLISHED: Consistent coverage over time
    • FADING: Decreasing coverage compared to history

Step 4: Calculate Momentum

For each topic:

momentum = (current_article_count - historical_average) / historical_average
  • Positive momentum: increasing coverage
  • Negative momentum: decreasing coverage
  • Zero/undefined: new topic or no history

Step 5: Detect Entity Clusters

Find entities that frequently co-occur in articles:

  • Build co-occurrence matrix
  • Identify pairs appearing together 2+ times
  • Group into clusters with total mention counts

Step 6: Store Theme for Future Tracking

Store each trend in memory for future comparisons.

Success Criteria

  • Topics correctly grouped by entity
  • Hot topics identified (multi-source coverage)
  • Trend status accurately reflects momentum
  • Entity clusters reveal related concepts

Trend Status Priority (for sorting)

  1. BREAKING - Immediate, new, high-impact
  2. HOT - Multiple sources, active coverage
  3. RISING - Growing in coverage
  4. ESTABLISHED - Ongoing coverage
  5. FADING - Decreasing coverage