ppt-classify

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

ppt-classify — PPT 类型分类 + 章节骨架 + 立论路由

ppt-classify — PPT Type Classification + Chapter Skeleton + Argumentation Routing

何时触发

When to Trigger

适用
  • 用户描述了一个 PPT 任务("帮我做一份 xxx 的 PPT")但还没定立论方式
  • 用户在动手做 PPT 之前,想先想清楚整体架构
  • 团队讨论 PPT 方向时需要对齐"这是在做什么类型的表达"+"整体多少章"
不适用
  • 已经明确是某一类 PPT + 章节已定 → 直接用对应立论 skill 深入每章
  • 纯数据报告 / 纯文字文档 → 不是 PPT 的问题,换格式
Applicable Scenarios:
  • The user describes a PPT task ("Help me create a PPT about xxx") but hasn't decided on the argumentation approach yet
  • Before starting to make the PPT, the user wants to clarify the overall structure first
  • During team discussions on PPT direction, it's necessary to align on "what type of presentation this is" + "how many chapters there will be in total"
Inapplicable Scenarios:
  • The PPT type is already clear and chapters are finalized → Directly use the corresponding argumentation skill to dive into each chapter
  • Pure data reports / pure text documents → This is not a PPT-related issue; switch formats

核心洞察:分类 + 骨架 = Step 0

Core Insight: Classification + Skeleton = Step 0

做 PPT 的第一个错误不是选错模板,是没分清你在做哪种 PPT、也没想过整体骨架
四种 PPT 的立论方式完全不同,对应的章节骨架也完全不同:
类型核心意图立论方式典型骨架形态典型场景
Pitch / Proposal说服观众接受某判断抛观点 → 捍卫痛点 → 方案 → 证据 → 路径 → CTA融资 / 立项 / 方案 / 汇报结论
Research / Investigation带观众走一遍调查提问题 → 设计路径 → 推结论悖论 → 被研究对象论证 → 多条检验路径 → 综合行业分析 / 竞品研究 / 投资
Teaching / How-to教观众做某事明确目标能力 → 拆步骤目标 → 前置 → 分步 → 验证 → 延伸培训 / 教程 / 方法论普及
Narrative / Storytelling讲一个故事主角 + 冲突 + 转折开场 → 冲突 → 对抗 → 突破 → 余韵复盘 / 品牌故事 / 案例分享
用 pitch 的方式做 research PPT → 你会变成鼓吹者。 用 research 的骨架做 pitch PPT → 观众听完不知道你想让他们做什么。 用 narrative 的骨架做 teaching PPT → 观众被故事打动但什么也没学会。
The first mistake in making a PPT is not choosing the wrong template, but not clarifying what type of PPT you're creating and not thinking about the overall skeleton.
The argumentation approaches for the four PPT types are completely different, and their corresponding chapter skeletons are also entirely different:
TypeCore ObjectiveArgumentation ApproachTypical Skeleton FormTypical Scenarios
Pitch / ProposalPersuade the audience to accept a certain judgmentPresent a viewpoint → Defend itPain Point → Solution → Evidence → Path → CTAFinancing / Project Initiation / Proposal / Conclusion Reporting
Research / InvestigationGuide the audience through an investigationRaise a question → Design a path → Derive a conclusionParadox → Argumentation of the Research Subject → Multiple Verification Paths → SynthesisIndustry Analysis / Competitor Research / Investment Research
Teaching / How-toTeach the audience to do somethingClarify target competence → Break down stepsObjective → Prerequisites → Step-by-Step → Verification → ExtensionTraining / Tutorial / Methodology Popularization
Narrative / StorytellingTell a storyProtagonist + Conflict + TwistOpening → Conflict → Confrontation → Breakthrough → AftermathProject Review / Brand Story / Case Sharing
Using a pitch-style approach for a research PPT → You'll come across as an advocate. Using a research-style skeleton for a pitch PPT → The audience won't know what you want them to do after listening. Using a narrative-style skeleton for a teaching PPT → The audience is moved by the story but learns nothing.

诊断流程

Diagnostic Process

逐层分层问(按顺序问,前一问命中就停止):
Ask questions layer by layer (in order; stop when the previous question hits):

Q1:这个 PPT 要让观众做出某个决策吗?

Q1: Does this PPT require the audience to make a specific decision?

  • → 极可能是 Pitch(融资、产品立项、方案批准都是决策驱动)
    • 进一步确认:结论是否已经定了?你在捍卫一个判断?→ Pitch 确认
    • 如果结论还没定,只是"想让观众参与决策" → 可能是 Research
  • → 继续 Q2
  • Yes → Most likely a Pitch (Financing, product initiation, proposal approval are all decision-driven)
    • Further confirmation: Is the conclusion already finalized? Are you defending a judgment? → Confirm Pitch
    • If the conclusion is not finalized and you just "want the audience to participate in the decision" → May be Research
  • No → Proceed to Q2

Q2:观众听完 PPT 需要会做某件具体的事吗?

Q2: Does the audience need to be able to do a specific task after listening to the PPT?

  • (能独立复现某个操作、应用某个方法) → Teaching
  • → 继续 Q3
  • Yes (Able to independently reproduce an operation or apply a method) → Teaching
  • No → Proceed to Q3

Q3:核心驱动力是"具体事件经过"还是"抽象论证"?

Q3: Is the core driving force "specific event process" or "abstract argumentation"?

  • 事件经过(有主角、有时间线、有情感弧线) → Narrative
  • 抽象论证(有框架、有证据链、有维度分析) → Research
  • Event process (Has protagonist, timeline, emotional arc) → Narrative
  • Abstract argumentation (Has framework, evidence chain, dimensional analysis) → Research

Q4:边界情况

Q4: Edge Cases

  • 混合型:标注主类型 + 次类型。主类型决定立论 skill + 骨架形态,次类型决定辅助手法
  • 用户不确定:列出最像的两类,描述各自的立论后果 + 骨架差异,让用户选
  • 根本不该做 PPT:如纯数据查询、纯 FAQ → 建议换格式
  • Hybrid type: Mark the main type + secondary type. The main type determines the argumentation skill + skeleton form, while the secondary type determines auxiliary techniques
  • User is uncertain: List the two most similar types, describe the argumentation consequences and skeleton differences for each, and let the user choose
  • Should not make a PPT at all: Such as pure data query, pure FAQ → Suggest switching formats

输出格式 · v0.2.0 新增「章节骨架建议」

Output Format · v0.2.0 New Addition: Chapter Skeleton Suggestions

完成诊断后,输出如下结构(必须包含四个 section):
【PPT 类型判定】
主类型:Research
次类型:(无 / 或注明)
判定依据:用户想带观众走一遍"NVIDIA 护城河是否削弱"的调查,
          结论不预设,观点在多条路径汇合后产生。

【推荐立论 skill】
主路径:ppt-research-setup(研究型三段式 + specificity 诊断 +
        本 skill 输出的章节骨架作为输入)
辅助:完成后可用 ppt-narrative-review 做张力审稿

【章节骨架建议】
研究型典型骨架(已针对"NVIDIA 护城河"personalize):
  章 1 · 反共识悖论     | 翻倍悖论:$500B → $1T · 21 个月
  章 2 · 被研究对象论证 | Jensen 的四步推理链(需求 / Token / TCO / 资金)
  章 3 · 检验路径 1     | 需求真爆炸?三行业 Token 落地 + 物理极限
  章 4 · 检验路径 2     | Token 经济 + NVIDIA 身份换皮
  章 5 · 检验路径 3     | TCO 护城河 + 中国替代链威胁
  章 6 · 综合判决       | 四步结论 + 3 个重估信号

建议章节数:5-7 章(研究型的适配范围)
建议单集长度:如 5 min 视频配套,每章约 30-60 秒;如 20 min 深度,可扩到每章 2-3 分钟

【下一步】
运行 ppt-research-setup,把上面的章节骨架传入,让它把每章填成
具体的推理链 + 证据锚点。
After completing the diagnosis, output the following structure (must include four sections):
【PPT 类型判定】
主类型:Research
次类型:(无 / 或注明)
判定依据:用户想带观众走一遍"NVIDIA 护城河是否削弱"的调查,
          结论不预设,观点在多条路径汇合后产生。

【推荐立论 skill】
主路径:ppt-research-setup(研究型三段式 + specificity 诊断 +
        本 skill 输出的章节骨架作为输入)
辅助:完成后可用 ppt-narrative-review 做张力审稿

【章节骨架建议】
研究型典型骨架(已针对"NVIDIA 护城河"personalize):
  章 1 · 反共识悖论     | 翻倍悖论:$500B → $1T · 21 个月
  章 2 · 被研究对象论证 | Jensen 的四步推理链(需求 / Token / TCO / 资金)
  章 3 · 检验路径 1     | 需求真爆炸?三行业 Token 落地 + 物理极限
  章 4 · 检验路径 2     | Token 经济 + NVIDIA 身份换皮
  章 5 · 检验路径 3     | TCO 护城河 + 中国替代链威胁
  章 6 · 综合判决       | 四步结论 + 3 个重估信号

建议章节数:5-7 章(研究型的适配范围)
建议单集长度:如 5 min 视频配套,每章约 30-60 秒;如 20 min 深度,可扩到每章 2-3 分钟

【下一步】
运行 ppt-research-setup,把上面的章节骨架传入,让它把每章填成
具体的推理链 + 证据锚点。

四类型的规范骨架(reference)

Canonical Skeletons for Four Types (Reference)

这是 skill 内部查表用的 canonical skeleton,personalize 时按具体题目替换章标题
This is the canonical skeleton used for internal lookup in the skill. Replace chapter titles with specific topics when personalizing:

Pitch / Proposal(5-7 章)

Pitch / Proposal (5-7 chapters)

  1. 痛点 / 现状(观众为什么该关心)
  2. 解决方案(我们提出什么)
  3. 为什么是我们(差异化 / 团队 / 资产)
  4. 牵引力 / 证据(用户 / 订单 / 实验数据)
  5. 商业模式 / 路径(钱从哪来 / 怎么 scale)
  6. (可选)风险缓释
  7. CTA / Ask(观众走出会议时要做什么)
  1. Pain Point / Current Situation (Why the audience should care)
  2. Solution (What we propose)
  3. Why Us (Differentiation / Team / Assets)
  4. Traction / Evidence (Users / Orders / Experimental Data)
  5. Business Model / Path (Where the money comes from / How to scale)
  6. (Optional) Risk Mitigation
  7. CTA / Ask (What the audience should do when leaving the meeting)

Research / Investigation(5-7 章)

Research / Investigation (5-7 chapters)

  1. 反共识悖论(研究动机:为什么这个问题不是显然的)
  2. 被研究对象的论证(先完整说清,不急着反驳) 3-5. 检验路径 1/2/3(每条独立、可验证、能推出证据)
  3. 综合判决 + 监测框架(结论 + 未来重审信号)
  1. Counterconsensus Paradox (Research motivation: Why this question is not obvious)
  2. Argumentation of the Research Subject (Explain it fully first; don't rush to refute) 3-5. Verification Path 1/2/3 (Each is independent, verifiable, and can produce evidence)
  3. Synthetic Judgment + Monitoring Framework (Conclusion + Future re-review signals)

Teaching / How-to(4-6 章)

Teaching / How-to (4-6 chapters)

  1. 学完能做什么(目标能力 / 验收标准)
  2. 前置条件(你需要哪些基础 / 准备什么工具) 3-5. Step-by-step(可按复杂度拆成 2-4 章)
  3. 常见陷阱 + 如何验证 + 延伸学习
  1. What You Can Do After Learning (Target competence / Acceptance criteria)
  2. Prerequisites (What basics you need / What tools to prepare) 3-5. Step-by-step (Can be split into 2-4 chapters based on complexity)
  3. Common Pitfalls + How to Verify + Extended Learning

Narrative / Storytelling(5 章 · 三幕式)

Narrative / Storytelling (5 chapters · Three-Act Structure)

  1. 开场:主角 + 世界设定
  2. 转折 1:冲突 / 触发事件
  3. 中段:对抗升级 / 代价
  4. 转折 2:突破 / 顿悟 / 关键决策
  5. 结尾:新常态 + 余韵 / 启示
  1. Opening: Protagonist + World Setting
  2. Twist 1: Conflict / Trigger Event
  3. Middle Act: Escalating Confrontation / Cost
  4. Twist 2: Breakthrough / Epiphany / Critical Decision
  5. Ending: New Normal + Aftermath / Inspiration

Personalize 规则

Personalization Rules

不要直接吐上面的 canonical skeleton 原文。按用户的具体题目:
  1. 识别题目里的关键概念(如 "NVIDIA 护城河"、"Claude Code 教程")
  2. 把 canonical 骨架的抽象 slot 填成具体章节标题
  3. 检查章节数是否匹配用户的时长预期
  4. 如果混合型(主 X + 次 Y),在骨架里嵌入 1-2 个次类型的章节(例:主 Research + 次 Narrative → 在某一条检验路径里加一个"关键个案"子章)
Do not directly output the original canonical skeleton above. For the user's specific topic:
  1. Identify key concepts in the topic (e.g., "NVIDIA Moat", "Claude Code Tutorial")
  2. Replace the abstract slots in the canonical skeleton with specific chapter titles
  3. Check if the number of chapters matches the user's expected duration
  4. If it's a hybrid type (Main X + Secondary Y), embed 1-2 chapters of the secondary type in the skeleton (Example: Main Research + Secondary Narrative → Add a "Key Case" sub-chapter in one of the verification paths)

快速参考:四类常见场景映射

Quick Reference: Mapping of Four Common Scenarios

用户描述判定骨架示例
"帮我做融资 PPT"Pitch痛点 → 方案 → 牵引 → 钱 → CTA
"帮我做一份 xx 行业的调研 PPT"Research悖论 → 对象论证 → 3 检验 → 综合
"帮我做一份 xx 产品的使用教程"Teaching目标 → 前置 → 3 步骤 → 验证
"帮我做一份项目复盘 PPT"Narrative开场 → 冲突 → 对抗 → 突破 → 余韵
"帮我做 PPT 汇报本季度 KPI"主 Pitch + 次 NarrativePitch 主骨架 + 开场用一个具体案例钩
"帮我做一份 NVIDIA 护城河分析"Research(见上面输出格式示例)
"帮我做一份 Claude Code 教程"Teaching能做什么 → 装 CLI → 基础命令 → 进阶 → 验证
"帮我做一份方案 PPT 给客户"主 Pitch + 次 ResearchPitch 骨架 + 论据章借 research 的证据展开
User DescriptionClassificationSkeleton Example
"Help me create a financing PPT"PitchPain Point → Solution → Traction → Funding → CTA
"Help me create a research PPT about xx industry"ResearchParadox → Subject Argumentation → 3 Verifications → Synthesis
"Help me create a usage tutorial PPT for xx product"TeachingObjective → Prerequisites → 3 Steps → Verification
"Help me create a project review PPT"NarrativeOpening → Conflict → Confrontation → Breakthrough → Aftermath
"Help me create a PPT to report this quarter's KPI"Main Pitch + Secondary NarrativeMain Pitch skeleton + Hook with a specific case in the opening
"Help me create an analysis of NVIDIA's moat"Research(See the example in the Output Format section above)
"Help me create a Claude Code tutorial"TeachingWhat you can do → Install CLI → Basic Commands → Advanced → Verification
"Help me create a proposal PPT for clients"Main Pitch + Secondary ResearchPitch skeleton + Use research-style evidence organization in the argument chapter

混合型骨架处理

Handling Hybrid Skeletons

主类型决定骨架形态。次类型在某一个章节里点缀,不改变整体骨架:
  • 主 Research + 次 Narrative:在 "检验路径 X" 的一个小节里插入个案故事
  • 主 Pitch + 次 Research:论据章(第 4 章)内部用研究型坐标系组织证据
  • 主 Teaching + 次 Narrative:开场痛点章改成"一个踩坑故事"
  • 主 Narrative + 次 Research:转折点(第 2/4 章)用反直觉数据作触发
The main type determines the skeleton form. The secondary type is added as an embellishment in one chapter without changing the overall skeleton:
  • Main Research + Secondary Narrative: Insert a case story in a subsection of "Verification Path X"
  • Main Pitch + Secondary Research: Organize evidence using a research-style framework inside the argument chapter (Chapter 4)
  • Main Teaching + Secondary Narrative: Change the opening pain point chapter to "A Pitfall Story"
  • Main Narrative + Secondary Research: Use counterintuitive data as the trigger for turning points (Chapters 2/4)

Claude Integration

Claude Integration

当用户的请求触发本 skill:
  1. 如果用户描述足够清晰:直接按 Q1-Q3 链路诊断,给出判定 + personalize 骨架
  2. 如果信息不足:用
    AskUserQuestion
    逐层追问。必问三项:题目 · 素材 / 一手来源 · 目标观众 · (可选)时长预期
  3. 输出骨架时一定要 personalize:canonical 骨架是底稿,成品必须贴具体题目
  4. 不要做立论:本 skill 只给类型 + 骨架 + 下一步 skill 名。具体每章的推理链由
    ppt-research-setup
    等后继 skill 填
When the user's request triggers this skill:
  1. If the user's description is clear enough: Directly diagnose using the Q1-Q3 chain, and provide the classification + personalized skeleton
  2. If information is insufficient: Use
    AskUserQuestion
    to inquire layer by layer. Must ask three items: Topic · Materials / First-hand Sources · Target Audience · (Optional) Duration Expectation
  3. Must personalize when outputting the skeleton: The canonical skeleton is a draft; the final product must be tailored to the specific topic
  4. Do not handle argumentation: This skill only provides the type + skeleton + name of the next skill. The specific reasoning chain for each chapter is filled by downstream skills like
    ppt-research-setup

与下游 skill 的契约

Contract with Downstream Skills

ppt-research-setup
等下游 skill 会把本 skill 的骨架当输入
  • ppt-classify
    输出章节名 + 章节角色 · strategic level
  • ppt-research-setup
    把每章填成 Section 1 (反共识悖论具体措辞) + Section 2 (推理链) + Section 3 (检验路径具体锚点) · tactical level
两层职责不重叠:classify 管"这片子有哪些章",research-setup 管"每章具体讲什么逻辑链"。
Downstream skills like
ppt-research-setup
will take the skeleton output by this skill as input:
  • ppt-classify
    outputs chapter names + chapter roles · strategic level
  • ppt-research-setup
    fills each chapter with Section 1 (Specific wording of counterconsensus paradox) + Section 2 (Reasoning chain) + Section 3 (Specific anchor points for verification paths) · tactical level
The two layers of responsibilities do not overlap: classify manages "what chapters this presentation has", while research-setup manages "what logical chain each chapter specifically covers"

设计来源

Design Origin

本 skill 从 EP10《AI 做 PPT》实验 A 提炼。实验 A 发现把 pitch 型立论 skill 套到 research 型 PPT 会把研究者推成鼓吹者——分类是做 PPT 的真正零步。
v0.2.0 变更:EP10 正式录屏前发现原 v0.1.0 只给类型标签不给骨架,导致用户从"知道是哪种 PPT"到"知道要有几章" 之间有断层。v0.2.0 把骨架建议前移到 classify,让 Step 0 真正能作为"整体定调"来用——给策略 + 给骨架,下游 research-setup 只管填充每章细节。
详见:
notes/experiment-a-ceo-review.md
和《PPT 演讲的分类和工作场景应用》。
This skill is refined from Experiment A of EP10《AI Makes PPT》. Experiment A found that applying pitch-style argumentation skills to research-type PPTs would turn researchers into advocates — classification is the true Step 0 in making PPTs.
v0.2.0 Changes: Before the official recording of EP10, it was found that the original v0.1.0 only provided type labels without skeletons, leading to a gap between users "knowing what type of PPT it is" and "knowing how many chapters there should be". v0.2.0 moves the skeleton suggestion to the classify step, allowing Step 0 to truly serve as "overall positioning" — providing strategy + skeleton, while downstream research-setup only focuses on filling in the details of each chapter.
For details:
notes/experiment-a-ceo-review.md
and 《Classification of PPT Presentations and Application in Work Scenarios》.