teachable-figures

改变一个条件，看清一个机制。Change one condition. See one mechanism.

一个给 Claude Code、Codex 等 agent 用的 skill。它把一句难懂的结论变成离线的互动图：学习者先预测，再动手改变一个原因，观察读数，最后用自己的话解释。

An agent skill for Claude Code, Codex, and others. It turns a hard-to-grasp claim into an offline interactive figure: the learner predicts, changes one cause, watches the readout, then explains it in their own words.

看它如何工作See how it works → 安装说明 Install guide
Claude Code
git clone https://github.com/ywEdAi/teachable-figures \
  ~/.claude/skills/teachable-figures
  1. 01
    预测Predict先选一个答案，其中一个是常见误解。Commit to an answer; one option is a common misconception.
  2. 02
    动手Try只改变一个原因，图和读数来自同一个模型。Change one cause; the figure and readouts share one model.
  3. 03
    观察Observe读数让结果可以对照你的预测。Readouts let you check the result against your prediction.
  4. 04
    解释与迁移Explain & transfer用自己的话说清楚，再预测一个新情形。Say why in your own words, then predict a new case.
先从这里开始Start here

这个 skill 先画图吗？Does this skill start by drawing?

用这个 skill 讲它自己：交给它几个请求，看哪些会被拒绝；再拿掉其中一步，看哪些失败在截图里根本看不出来。

The skill, explained with itself: hand it requests and see which it declines, then remove a step and see which failures a screenshot never shows.

打开 →Open →

六个完整示例Six complete lessons

都是单个 HTML 文件，离线可用，写明了模型的假设与来源。Each is one offline HTML file that states its model's assumptions and sources.