Agent-Skills.md

Agent Skills: Content Digest

>

UncategorizedID: zephyrwang6/myskill/content-digest

Repository

zephyrwang6/myskill
zephyrwang6
3814

Install this agent skill to your local

pnpm dlx add-skill https://github.com/zephyrwang6/myskill/tree/HEAD/content-digest

Skill Files

Browse the full folder contents for content-digest.

Download Skill

Loading file tree…

content-digest/SKILL.md

Skill Metadata

Name
content-digest
Description
>

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:

# MMDD：[嘉宾名] X [栏目名]：[一句话核心观点]

今天看到 [嘉宾名] 去了 [栏目名] 的播客。

[嘉宾名] [2-3句话介绍嘉宾背景和核心成就，用具体数据]。

这期播客总共录了 [时长]，[嘉宾名] 谈到了 [N] 个有趣的观点：

1、[观点标题/关键词]。[完整的逻辑阐述，包含推理过程，2-4句话]

2、[观点标题/关键词]。[完整的逻辑阐述，包含推理过程，2-4句话]

3、[观点标题/关键词]。[完整的逻辑阐述，包含推理过程，2-4句话]

...
[Continue with 10-15 total points]

---

标题格式 (重要):

  • 格式：# MMDD：嘉宾名 X 栏目名：一句话观点
  • 示例：# 0130：Peter Steinberger X The Pragmatic Engineer：一天600次提交，代码比以前更好
  • 一句话观点要抓住最反直觉或最有冲击力的点

开头格式 (重要):

  • 第一句：简单说看了什么 今天看到 [嘉宾] 去了 [栏目] 的播客。
  • 第二段：2-3句介绍嘉宾背景，必须有具体数据（数字、公司名、产品名）
  • 第三句：过渡句 这期播客总共录了 [时长]，[嘉宾] 谈到了 [N] 个有趣的观点：
  • 如果不知道时长，可以写"将近两小时"或省略时长

List Format Rules (重要):

  • 使用数字+顿号格式 (1、 2、 3、...)
    • 注意是顿号「、」不是点号「.」
  • 每条是完整段落，包含：
    • 观点/结论（第一句）
    • 逻辑推理/因果解释（后续句子）
    • 具体例子或数据（如有）
  • 每条 2-4 句话，50-150 字
  • 观点要有深度：
    • ✅ "AI 应用创业者不会相信 AGI。逻辑很简单，如果真信，就不应该做 AI 应用创业。AGI 如果存在，创业就只剩一件事：去做有机会达成 AGI 的模型。"
    • ❌ "AI 应用创业者不相信 AGI"（太简单，没有逻辑推理）
  • 保持 10-15 条（不是 8-12 条）

Key principles:

  • 标题要有冲击力，抓住最反直觉的观点
  • 开头简洁，快速进入正题
  • 每条观点要有完整的逻辑链条
  • 观点之间可以有递进或对比关系
  • Include specific data when available (numbers, names, percentages)
  • Pass the non-obvious test (would informed readers NOT already know this?)
  • End with separator line ---

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 separate summary - it's a narrative expansion of the SHORT-form insights with story arc and analytical depth.

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

IMPORTANT: Choose the right style based on content type

For interview/podcast/dialogue content → Use Style B (对话式访谈) For solo speech/article/essay → Use Style A (叙事性文章)

Style A Structure (叙事性文章):

[Compelling Title - derived from core insight]

[Opening: Set the scene using one of the 10-15 insights]

### [Section 1: Background]
"[Key quote]"
[Context - connect to 2-3 of your curated insights]

### [Section 2: Main Content]
[Narrative development - weave in 4-5 curated insights with quotes and analysis]

### [Section 3: Climax]
[Dramatic highlights - reveal most counterintuitive insight]

### [Section 4: Resolution/Turning Point]
"[Pivotal quote]"
[Significance - tie back to mental model or principle]

### [Epilogue: Reflection]
[What happened after / Ironic contrast - connect final insights]

[Optional: Source attribution]

Key principles (Style A):

  • Compelling title derived from your deepest insight
  • Clear section headers for navigation
  • Each section develops 2-4 of your 10-15 curated insights with narrative and quotes
  • Integrate direct quotes naturally - use them to prove your curated insights
  • Build narrative arc: setup → development → climax → resolution
  • Use dramatic irony when relevant ("他不知道的是..." / "He didn't know that...")
  • Include specific details (names, numbers, dates) from your curated list
  • Deep analytical layer - weave in your Stage 2 filtered insights:
    • Why specific choices reveal broader strategic principles
    • How contradictions or tensions expose underlying philosophies
    • What the subject's evolution teaches about the domain
    • Connections between micro-decisions and macro-outcomes
  • 2000-3500+ characters to properly develop 10-15 deep insights

Style B Structure (对话式访谈):

[前言/导语 - 编辑者视角]
断断续续，终于看完了...[个人感受]
干货很多。[嘉宾]可能是...[定位评价]
[为什么值得关注]
我今天不忙，把这次访谈全文精编出来，供大家学习。赠人玫瑰，手有余香。
[可选：节日祝福]
下面是 YouTube/播客链接：[链接]

#01 [主题标题 - 简短有力]
主持人：[问题]
嘉宾：[回答 - 保留对话感]

[编辑补充：数据解读、背景、个人观点]
[可用："我觉得这个点真的太重要了" "这太有意思了"]

主持人：[追问]
嘉宾：[深入回答]

[继续分析]
[可选：插入相关文章链接]

#02 [第二主题]
主持人：[新话题]
嘉宾：[回答]

[编辑解读]
...

#03-#0N [按主题继续]
...

Key principles (Style B):

  • 口语化开场："断断续续看完" "干货很多" 体现真实感
  • 编号主题：用 #01 #02 等清晰分段，主题标题直白有力
  • 保留对话：60-70%保持"主持人：""嘉宾："格式
  • 编辑介入：20-30%加入编辑分析、补充、个人反应
  • 口语化表达："太离谱了" "我觉得" "说实话" "天哪" "完全是这样"
  • 具体数据：必须保留数字、人名、公司名
  • 补充链接：适时插入"文章链接：..." 延伸阅读
  • 人情味结尾："赠人玫瑰，手有余香"
  • 主题重组：不按时间线，按话题逻辑重新组织
  • 3000-8000+ 字（根据访谈长度）

6. Quality Check

Before delivering, verify you followed the four-stage process:

Stage 1 verification:

  • [ ] Extracted 50+ viewpoints from source material (can be implicit - doesn't need to be shown to user)

Stage 2 verification:

  • [ ] Filtered for non-consensus, counterintuitive, and deep insights
  • [ ] Identified personal/private expressions and interesting trivia
  • [ ] Marked mental models and second-order insights

Stage 3 & 4 verification:

  • [ ] Selected 10-15 most profound insights for final output
  • [ ] Short version uses ALL 10-15 curated insights
  • [ ] Long version develops the SAME 10-15 insights with narrative

Depth check - Each of the 10-15 insights must:

  • [ ] Pass the "non-obvious test": Would an informed reader already know this?
  • [ ] Reveal a mental model, framework, or underlying principle
  • [ ] Challenge conventional thinking OR expose interesting trivia
  • [ ] Connect ideas in an unexpected way OR show second-order effects

Quality verification:

  • [ ] Quotes are accurate and attributed
  • [ ] No editorializing beyond source material
  • [ ] Writing is engaging, not robotic
  • [ ] Both versions can stand alone
  • [ ] Numbers/facts are specific, not vague
  • [ ] The "so what?" is clear to readers
  • [ ] Every takeaway reveals WHY it matters, not just WHAT happened
  • [ ] Short and long versions share the same insight foundation

7. Deliver Output

文件保存规范：

  • 保存位置：/Users/ugreen/Documents/obsidian/每日播客/
  • 文件命名：MMDD-主题关键词.md
    • MMDD 为当天日期（如 0109 表示 1 月 9 日）
    • 主题关键词 为 2-6 个字的内容概括（如 Lovable增长策略、AI编程工具）
    • 示例：0109-Lovable增长策略.md、0108-睡眠科学.md
  • 自动保存：生成内容后，使用 Write 工具将完整内容保存到上述路径

格式规范：

  • 行距：段落内不留空行，段落之间留一个空行
  • 文档标题：使用 # MMDD：嘉宾名 X 栏目名：一句话观点 格式
  • 长文标题：使用 # 精华片段

结构模板：

# MMDD：[嘉宾名] X [栏目名]：[一句话核心观点]

今天看到 [嘉宾名] 去了 [栏目名] 的播客。

[嘉宾名] [2-3句话介绍嘉宾背景，用具体数据]。

这期播客总共录了 [时长]，[嘉宾名] 谈到了 [N] 个有趣的观点：

1、[观点]。[逻辑推理]

2、[观点]。[逻辑推理]

...（共 10-15 条）

---

# 精华片段

[长文内容 - Style B 对话式访谈格式]

---

[结束语]

开场介绍模板（新格式）：

# 0130：Peter Steinberger X The Pragmatic Engineer：一天600次提交，代码比以前更好

今天看到 Peter Steinberger 去了 The Pragmatic Engineer 的播客。

Peter Steinberger 创建了 PSPDF kit——一个被超过 10 亿设备使用的 PDF 框架，后来经历严重职业倦怠卖掉股份消失了 3 年。2024 年回归后，他用完全不同的方式创建了 Clawdbot，一周从 100 星涨到 3300 星。

这期播客总共录了将近两小时，Peter 谈到了 11 个有趣的观点：

观点列表示例：

1、AI 应用创业者不会相信 AGI。逻辑很简单，如果真信，就不应该做 AI 应用创业。AGI 如果存在，创业就只剩一件事：去做有机会达成 AGI 的模型。

2、创业的产品定位，关键点之一是：要做工作流的上游。上游的产品，不容易被下游的产品吃掉，反过来，发展到一定阶段，上游可以做下游的事。

3、Code Reviews 已死。对 prompts 的兴趣比代码更大，PR 应该叫"Prompt Requests"。因为 prompt 给的信号更高——你是怎么得到这个解决方案的？问了什么？有多少引导？

结束语模板：

---

[嘉宾名字]凭[具体贡献]，[产生的影响]。

用他的话收个尾："[引用一句有力的话]"。[最后一句个人感受或升华]。

YouTube 链接：[链接]

If only one format was requested, still include opening and closing sections.

Core Principles

Extract Strategically

What to extract:

  • Deep, counterintuitive insights - not surface observations:
    • Mental models and frameworks that drive decision-making
    • Paradoxes and tensions that reveal underlying principles
    • Second-order effects and non-obvious consequences
    • Patterns that connect specific tactics to strategic outcomes
  • Surprising insights that challenge common wisdom
  • Practical wisdom with WHY - not just "do X" but "X reveals principle Y"
  • Memorable quotes capturing big ideas or philosophical stances
  • Turning points and paradigm shifts
  • Human moments (vulnerability, humor, authenticity) that reveal character
  • Contextual ironies (what they didn't know then vs. now)

What to avoid:

  • Linear summarization without insight
  • Including everything (be selective)
  • Stating the obvious - if a reasonably informed reader would already know it, dig deeper
  • Surface-level descriptions without explaining WHY it matters
  • Losing the human voice
  • Adding information not in source
  • AI-style generic phrasing
  • Shallow takeaways - "X did Y" without revealing what principle or framework this demonstrates

Voice & Tone

  • Conversational but insightful: Like explaining to a smart friend
  • Show, don't tell: Use quotes to prove points
  • Respect the source: Don't editorialize or distort
  • Find the story: Every piece has a narrative arc

Be Token-Efficient

This skill focuses on creative transformation, not code execution. The writing process happens in-context without requiring scripts.

Resources

references/style-guide.md

Detailed writing guidelines for both short-form and long-form styles, including:

  • Structure patterns
  • Key characteristics
  • Writing principles
  • What to avoid
  • Extraction strategies

Load this when you need detailed guidance on tone, structure, or style.

references/examples.md

Complete reference examples:

  • Short-form example: Boris Cherny's Claude Code workflow
  • Long-form example: Manus/Peak Ji interview article

Load this when you need concrete examples of the final output quality and style.

Notes

  • Both styles require full comprehension of source material - don't skim
  • Short-form emphasizes actionable takeaways
  • Long-form emphasizes narrative and character
  • Quotes must be accurate and in context
  • Works best with content that has inherent narrative or insight
  • Can combine with translation if source is in different language
  • Ideal for content creators repurposing long content for different platforms