Skills that make Claude tutor the way the best human teachers do — guiding you to the answer instead of handing it over, so you actually learn it.
AI tutors have one failure mode: they explain too well. A fluent answer feels like understanding — you nod, move on, and can't reproduce it a week later. Research on AI-assisted learning shows measurable skill decline in students who lean on generated answers.
These skills are engineered against exactly that. They make you do the thinking, and they're built on established education research — not vibes.
| Skill | What it is |
|---|---|
tutor |
Universal tutor for any subject. The shared engine below, plus a path mode for "I want to learn X." Used as the fallback when no specialist is installed. |
math-tutor |
Math, algebra through calculus, probability, and linear algebra — with a catalog of the specific misconceptions learners hold at each topic. |
programming-tutor |
Programming, language-agnostic. Built on the notional machine (most bugs are a wrong model of execution) and teaches debugging as a scientific method, not random trial-and-error. |
- A hint ladder, not an answer. One nudge per turn — diagnose, check the foundation, surface the strategy — and you produce the next step. Producing it is what builds the skill; receiving it doesn't.
- It reads the situation first. Stuck on a problem, wanting a concept explained, and checking finished work each demand a different response. Giving the wrong one is the most common way tutoring fails.
- No "does that make sense?" — a yes/no question that everyone answers "yes." Replaced with reconstruction and prediction: the only checks a confused learner can't fake with a nod.
- A misconception catalog per subject. Not generic tips — the exact wrong rules learners actually hold, each with a counterexample you compute yourself. Believe
(a+b)² = a²+b²? You work out(1+1)² = 4 ≠ 2and watch your own rule break. That sticks; being told you're wrong doesn't.
Grounded in Skemp's relational understanding, du Boulay's notional machine, PRIMM, and cognitive load theory — sources in each skill's references/.
- Download a skill folder as a ZIP — open it (e.g.
math-tutor), click the green Code button → Download ZIP. - In claude.ai: Settings → Customize → Skills → + → Upload a skill, then upload the ZIP.
- Done — Claude uses it automatically when you study.
Have an idea for a skill? Open an issue or a pull request.