这个 skill 先画图吗?Does this skill start by drawing?
teachable-figures 把一个知识点做成“先预测、再动手、后解释”的互动图。这一页本身也是用它做的:给它几个不同的请求,再拿掉其中一步,看看结果会怎样。
teachable-figures turns one idea into a “predict, try, explain” interactive figure. This page is built the same way: hand it a few different requests, then remove a step and see what breaks.
01预测Predict
02交给它一个请求Hand it a request
选一个请求,左边的流程会显示每一步做出的决定。至少试两个,其中一个最好是它不该做的。
Pick a request. The flow on the left shows the decision each step makes. Try at least two, ideally one it shouldn't build.
引导模式只问尚未明确的信息,至多五个问题,每题都带推荐答案。下面展示完整的五个话题示例,已知答案不再询问;说“直接生成”就跳过提问,按默认值做,并在交付时列出假设。Guided mode asks only about missing details, up to five questions, each with a recommended answer. Below is a full five-topic example; known answers are not asked again. Saying “just build it” skips the questions; it uses defaults and lists its assumptions on delivery.
03拿掉一步Remove a step
如果跳过其中一步,产出还“看起来完成了”吗?还能教会吗?逐个勾选,留意左下角的产出和读数。
If one step is skipped, does the output still look finished? Does it still teach? Tick them one at a time and watch the output and readouts.
04解释Explain
这个 skill 先决定学习者要改变什么、观察什么,然后才决定画什么。顺序是:
- 范围:没有可操作机制的请求(数据看板、整门课)在这里被挡下或改道。
- 交互模式:先问一轮,或按默认值直接做。
- 找机制:一个原因、一个结果、一个可能被推翻的误解。
- 选渲染:按主体选——示意图、iso-figure、Anatomy 机械套件,或来源网格。
- 教学规格:在画任何几何之前写好。
- 从一个模型构建:图和读数都只“观察”这个模型。
- 可操作,然后验证。
刚才拿掉的几步里,只有“选渲染方式”的失败能在截图里看出来。没有可改变的原因、没有预测、只能用鼠标、端点算错——这些情况下图都“看起来完成了”。所以最后一步是操作这张图来验证,而不只是看截图。
This skill decides what the learner will change and observe before it decides what to draw. The order is:
- Scope: requests with no mechanism to operate (dashboards, whole courses) are declined or redirected here.
- Interaction mode: ask one round of questions, or build from defaults.
- Mechanism: one cause, one effect, one misconception that can be disproved.
- Renderer: chosen by subject — schematic, iso-figure, the Anatomy mechanical kit, or source meshes.
- Teaching spec: written before any geometry.
- Build from one model: the picture and the readouts only observe it.
- Make it operable, then verify.
Of the steps you removed, only a skipped renderer choice shows up in a screenshot. With no cause to change, no prediction, mouse-only controls, or a wrong endpoint, the figure still looks finished. That is why the last step operates the figure instead of only looking at it.
“它先写 ___,再画 ___,因为一张只是看起来完成的图 ___。”“It writes ___ before it draws ___, because a figure that only looks finished ___.”
迁移:换一个新请求Transfer: a new request
这个模型的边界What this model leaves out
这里的流程是 SKILL.md 的简化版:每个请求的决定是预先写好的,用来说明规则,不是真的在调用模型。真实使用时,agent 会读你的材料和项目,再做这些判断。
“拿掉一步”的后果是典型失败,不是每次都会发生的必然结果。这个 skill 产出的是设计良好的图,它们还没有经过学习者实验验证。
The flow here simplifies SKILL.md: each request's decisions are written in advance to illustrate the rules; no model is called. In real use, the agent reads your material and project before making them.
The effects of removing a step are typical failures, not guaranteed outcomes. The skill produces well-designed figures; they have not been tested with learners.
离线可用。所有控件都是原生表单元素,可用 Tab、空格和方向键操作。Works offline. Every control is a native form element you can reach with Tab, Space, and the arrow keys.
流程:从请求到可教学的图Flow: from request to teachable figure