compose-digest
Compose a formatted news digest from processed articles and trends. Uses LLM to generate cohesive narrative summary with embedded citations. Formats for Discord readability.
📖 中文导读:Compose a formatted news digest from processed articles and trends. Uses LLM to generate cohesive narrative summary with embedded citations. Formats for Discord readability. 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: compose-digest description: > Compose a formatted news digest from processed articles and trends. Uses LLM to generate cohesive narrative summary with embedded citations. Formats for Discord readability. metadata: domain: news category: action mcp-servers: [] requires-approval: false confidence: 0.85 input:
- name: articles type: list[ProcessedArticle] description: Processed articles to include in digest
- name: trends type: list[TrendingTopic] description: Identified trends for the period
- name: period_hours type: int default: 12 description: Hours covered by this digest output:
- name: digest type: NewsDigest description: Complete formatted news digest
- name: discord_formatted type: str description: Digest formatted for Discord posting
Compose Digest
Create a formatted news digest from articles and trends.
Preconditions
- Articles processed with analyze-article skill
- Trends analyzed with analyze-trends skill
- LLM API available for summary generation
Actions
Step 1: Select Articles by Importance
Sort articles by importance score and select:
- All high importance (score >= 7)
- Top 5 medium importance (score 5-6)
- Fallback: If no notable articles, take top 5 by score
Step 2: Generate LLM Summary
Send selected articles to LLM with prompt:
Write a cohesive, professional summary of these news items:
1. Be written as flowing paragraphs, not bullet points
2. Embed source citations inline using markdown [Source](url)
3. Highlight the most important developments first
4. Group related news naturally in the narrative
5. Be concise but comprehensive
Write 2-4 paragraphs summarizing the key AI news.
Step 3: Parse and Clean Response
- Strip Qwen3 thinking tags (
<think>...</think>) - Clean up any formatting artifacts
- If LLM fails, use fallback bullet-point format
Step 4: Build NewsDigest Object
Create digest with:
- Unique ID:
digest-{uuid[:8]} - Created timestamp
- Period start/end based on
period_hours - Generated headline summary
- Top 5 trends
- Article count
- Sources used
Step 5: Format for Discord
Format digest with sections:
# AI News Digest - {date} ({Morning/Evening})
{LLM-generated summary with citations}
**Trending Topics:**
- {topic} (covered by {n} sources)
**Emerging Themes:**
- {rising topic}
---
*{n} articles from {n} sources*
Success Criteria
- Coherent summary generated
- Citations properly embedded
- Trends highlighted appropriately
- Format fits Discord constraints
Fallback Behavior
If LLM unavailable:
**Today's AI News Highlights:**
- [{title}]({url}) ({source}): {summary}...