A Claude Code skill that teaches one concept by building the smallest runnable project that demonstrates it — one layer at a time, stopping for your explicit approval before each new layer.
You learn by watching the example grow, not by being handed the finished thing.
/try-example cosine similarity for course recs
- Mental model first. It states the single idea the concept rests on, then asks you to restate it. Nothing gets built until you do.
- One layer per turn. It writes the smallest meaningful increment, runs it, and shows the real output.
- Approval gate. It explains the layer, asks a comprehension check, then stops. The next layer doesn't appear until you say "next".
- Done when it's convincing. It stops at a decent example — not a product. No frameworks, no polish the concept didn't need.
Stdlib-first, zero new dependencies unless the concept truly needs one, fewest files, each concept in its own isolated folder.
git clone https://github.com/jerryzhou196/try-example.git ~/Github/try-example
ln -s ~/Github/try-example ~/.claude/skills/try-exampleRestart Claude Code so it picks up the skill, then run /try-example <concept>.
MIT