Content Digest

Transform long-form content into compelling short-form and long-form narratives.

Overview

This skill converts lengthy content (YouTube videos, podcasts, interviews, articles) into two distinct formats:

1. Short-Form (短文): Social media-friendly summaries (300-800 characters) with numbered emoji lists (1️⃣2️⃣3️⃣)
2. Long-Form (长文): Narrative articles (1500-3000+ characters) with story arcs, section headers, and integrated quotes

Workflow

1. Obtain the Content

If user provides a URL:

• YouTube links: Use WebFetch or attempt to extract transcript
• Article URLs: Use WebFetch to retrieve content
• Podcast links: Fetch transcript if available

If user provides text:

• Read the full transcript or article text directly

If content is unclear:

• Ask: "Please provide the YouTube link, podcast transcript, or article you'd like me to transform."

2. Determine Output Format

If user specifies format:

• Proceed with their choice (short-form only, long-form only, or both)

If user does not specify:

• Ask: "Would you like: (1) Short-form only, (2) Long-form only, or (3) Both versions?"

Default behavior:

• Generate both versions to maximize value

3. Deep Analysis - Four-Stage Process

CRITICAL: Follow this systematic process to ensure depth

Stage 1: Extract All Viewpoints (50+ minimum)

Read the entire content thoroughly and extract ALL viewpoints, including:

• Explicit statements and opinions
• Implicit beliefs revealed through stories
• Decision-making rationales
• Observations about the industry/domain
• Personal experiences and lessons
• Counterexamples and contrasts
• Numbers, data points, specific examples

Goal: Create a comprehensive list of 50+ viewpoints before filtering. Don't judge quality yet - just extract everything.

Stage 2: Filter for Non-Consensus & Depth

From the 50+ viewpoints, identify and mark those that are:

• Non-consensus (非共识): Challenges industry conventional wisdom
• Personal/private insights (个人私下表达): Things people think but rarely say publicly
• Counterintuitive (反直觉): Surprises even informed readers
• Interesting trivia (有意思的冷知识): Specific details that reveal deeper patterns
• Mental models: Frameworks that explain decision-making
• Second-order insights: Not just "what" but "why this matters philosophically"
• Paradoxes and tensions: Contradictions that expose underlying principles

Goal: Flag the 20-30 viewpoints that pass the "non-obvious test" - would a smart, informed reader already know this?

Stage 3: Select Core Narrative Elements

Identify:

• Core narrative: What's the main story or theme?
• Memorable quotes: Direct quotes that capture big ideas or reveal character
• Turning points: Moments of realization or paradigm shifts
• Dramatic elements: Irony, contrast, or unexpected outcomes
• Specific details: Names, numbers, dates that prove the deeper point

Stage 4: Curate Final Insights

From the filtered viewpoints (Stage 2) and narrative elements (Stage 3):

• For short-form: Select 10-15 most profound, actionable insights
• For long-form: Use the same 10-15 insights as the foundation, then weave in narrative arc

4. Generate Short-Form Version

CRITICAL: Use ONLY the 10-15 curated insights from Stage 4

Consult style-guide.md for detailed guidelines. See examples.md for reference.

Structure:

[Hook: Who said what / What happened]

[1-2 profound core insights - must be non-consensus or counterintuitive]

[Transition phrase like "总结一下做个笔记👇"]

1. **关键词/小标题**：一句话洞察，简洁有力
2. **关键词/小标题**：一句话洞察，简洁有力
3. **关键词/小标题**：一句话洞察，简洁有力
...
[Continue with 8-12 total points]

---

List Format Rules (重要):

• 使用数字+点号格式 (1. 2. 3....)，不要用emoji数字（1️⃣2️⃣...）
  • 原因：emoji数字超过10后不美观，且占用更多字符
• 每条以加粗关键词开头：**咖啡因机制**：... 或 **关于午睡**：...
  • 关键词2-6个字，概括该条主题
  • 冒号后用一句话说清楚洞察
• 每条控制在1-2句话，不超过50字
  • 如果需要更多解释，拆成两条
• 具体数据放在句中，不要单独列出
  • ✅ "REM睡眠每减少5%，死亡风险增加13%"
  • ❌ "REM睡眠很重要。研究显示减少5%会增加13%死亡风险。"

Key principles:

• Start with attention-grabbing hook
• Highlight 1-2 most profound non-consensus insights that:
  • Challenge how readers think about the domain
  • Reveal underlying mental models or strategic frameworks
  • Connect seemingly unrelated ideas to expose patterns
• Keep list to 8-12 items (not 15+, quality over quantity)
• Each point must:
  • Start with bold keyword for scannability
  • Be one core idea - if you need "而且/另外", split into two items
  • Include specific data when available (numbers, names, percentages)
  • Pass the non-obvious test (would informed readers NOT already know this?)
• End with separator line ---
• Target: 600-1000 characters for the list portion

5. Generate Long-Form Version

CRITICAL: Build ENTIRELY on the same 10-15 curated insights from Stage 4

The long-form version is NOT a sepa

...