Back to Details

make-it-click

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Make It Click

让内容一目了然

Models now do more of the work. People spend more of their time checking and understanding that work. This skill makes the output easy to understand. It treats the format of an answer as a choice, not a default, and it moves up a ladder of formats when the content needs it.
Idea credit: Andrej Karpathy's public notes on making LLM outputs easier to understand.
如今，模型承担了更多工作，因此人们需要花更多时间检查和理解这些工作。该 skill 让输出变得容易理解。它将答案格式视为一种可选方案，而不是默认设置；当内容需要时，它会沿着格式阶梯逐级升级。
思路来源：Andrej Karpathy 关于让 LLM 输出更易理解的公开笔记。

The ladder

格式阶梯

Each step costs more to make and gives more understanding. Pick the lowest step that does the job, unless the user asks for a specific step.
StepFormatBest forReference
1Controlled writing (STE style)Instructions, explanations, summaries, anything read once
references/ste-writing.md
2DiagramStructure, flow, relationships, comparisons, timelines
references/diagrams.md
3Interactive HTML pageContent with several layers, data to explore, parameters to change
references/html-explainers.md
4Narrated explainer videoConcepts that change over time, math, intuition building, teaching
references/explainer-videos.md
Read only the reference file for the step you use. Read 
ste-writing.md
for every step, because the words in diagrams, pages, and narration follow the same rules.
每一级制作成本更高，但带来的理解效果也更强。除非用户指定某一级，否则选择能够完成任务的最低一级。
步骤格式最适合的内容参考资料
1受控写作（STE 风格）说明、解释、摘要，以及只需阅读一次的内容
references/ste-writing.md
2图示结构、流程、关系、比较、时间线
references/diagrams.md
3交互式 HTML 页面包含多个层次的内容、可探索的数据、可调整的参数
references/html-explainers.md
4带旁白的讲解视频随时间变化的概念、数学、直觉构建、教学
references/explainer-videos.md
仅阅读你所用级别对应的参考文件。每一种级别都要阅读 
ste-writing.md
，因为图示、页面和旁白中的文字都遵循相同规则。

How to choose

如何选择

  1. If the user names a format, use that format.
  2. If the content has a shape (parts that connect, steps that branch, values that change), go to step 2 or higher. Prose hides shape.
  3. If the user will want to poke at it (change a value, expand a detail, filter a list), go to step 3.
  4. If the idea depends on motion or build-up (a transform, an algorithm running, a proof step by step), offer step 4. Videos take time and may need API keys, so confirm before you build one unless the user asked for it.
  5. Otherwise, use step 1.
When in doubt, give a short step-1 answer and offer the next step in one line.
  1. 如果用户指定了格式，就使用该格式。
  2. 如果内容具有某种结构形态（相互连接的部分、会分叉的步骤、会变化的值），就使用第 2 级或更高。文字会隐藏这种形态。
  3. 如果用户想进一步操作内容（更改数值、展开细节、筛选列表），就使用第 3 级。
  4. 如果概念依赖动态过程或逐步构建（某个变换、算法运行、逐步证明），就提供第 4 级方案。视频制作耗时，并且可能需要 API keys，因此除非用户已经要求，否则制作前应先确认。
  5. 否则，使用第 1 级。
如有疑问，先给出简短的第 1 级答案，并用一行提供下一级选项。

STE strength levels

STE 简化程度

Users can ask for different strengths. Default to "80%" when the user does not say.
  • Strict: follow every rule in 
    ste-writing.md
    , including the approved-word list.
  • 80% (default): follow the sentence, verb, and structure rules. Allow normal words when no simple word exists, and allow a few longer sentences to keep the text natural.
  • Light: short sentences, active voice, one idea per sentence. No other rules.
用户可以要求不同的简化程度。用户未说明时，默认使用“80%”。
  • 严格：遵循 
    ste-writing.md
    中的每条规则，包括批准词汇列表。
  • 80%（默认）：遵循句子、动词和结构规则。如果没有简单词汇可替代，允许使用常规词汇；为了让文字自然，允许少量较长句子。
  • 轻度：使用短句和主动语态，每句只表达一个想法，不执行其他规则。

Oversight mode

监督模式

Use this mode when the user must check work that a model or agent did (code changes, a research report, a long log, a data pipeline). Make the output answer these questions, in this order:
  1. What changed or what was found? (one sentence)
  2. Why? (the reason or the evidence)
  3. What is the risk? (what can break, what is not certain)
  4. What must the human check? (a short list of specific things to look at)
Put uncertainty in the open. Mark claims that you did not verify. A diagram of "before" and "after" is often the fastest way to show a change.
当用户必须检查模型或 Agent 完成的工作时（代码变更、研究报告、长日志、数据管道），使用此模式。让输出按以下顺序回答这些问题：
  1. 发生了什么变化或发现了什么？（用一句话）
  2. 为什么？（原因或证据）
  3. 风险是什么？（可能出什么问题，或哪些内容尚不确定）
  4. 人类必须检查什么？（列出需要查看的具体内容）
明确呈现不确定性。标记未经验证的主张。绘制“之前”和“之后”的对比图，通常是展示变更最快的方式。

Disposable artifacts

一次性产物

Pages and videos made with this skill are for one person and one question. Do not over-engineer them. Make them self-contained (one HTML file, one video file), make them fast, and do not add features the user did not ask for. It is acceptable to discard them after use.
用此 skill 制作的页面和视频只服务于一个人提出的一个问题。不要过度设计。使其保持自包含（一个 HTML 文件、一个视频文件），速度要快，并且不要添加用户未要求的功能。使用后可以丢弃。

Quality check before you deliver

交付前的质量检查

  • Can the user get the main point in ten seconds?
  • Does each sentence, label, or scene carry one idea?
  • Is the same term used for the same thing everywhere?
  • For steps 2-4: do the words also pass the step-1 rules?
  • Optional: run 
    python scripts/ste_check.py <file>
    on written text to flag long sentences, -ing forms, passive voice, and common unapproved words.
  • 用户能否在十秒内抓住要点？
  • 每个句子、标签或场景是否只承载一个想法？
  • 同一概念是否在所有位置使用同一个术语？
  • 对于第 2-4 级：其中的文字是否也通过第 1 级规则？
  • 可选：对书面文本运行 
    python scripts/ste_check.py <file>
    ，标记过长句子、-ing 形式、被动语态和常见未批准词汇。