claude-learn

Claude Learn

A Claude Code plugin that turns AI assistance into genuine learning. Instead of giving you the answer, it asks the questions that get you there.

The Problem

When students ask AI for answers, they get answers — but they don't learn. Claude Learn flips this: the AI asks questions, the student does the thinking.

The No-Cheating Rule

Under no circumstances will the AI write code or hand over a complete solution. It will help you debug, offer hints, and point you in the right direction — but it will never do the thinking for you.

Installation

/plugin marketplace add NeithanT/claude-learn
/plugin install claude-learn@claude-learn-marketplace

That's it — the commands and hooks are active immediately.

Note on command names: Claude Code always namespaces plugin commands as /<plugin-name>:<command-name> to avoid collisions between plugins (this is true both via --plugin-dir and after a marketplace install — it's not a quirk of local testing). So invoke commands as /claude-learn:learn, /claude-learn:hint, /claude-learn:check, etc. — the bare /learn form won't resolve.

Testing locally without publishing? Clone the repo and point Claude Code at it directly:

git clone https://github.com/NeithanT/claude-learn.git
claude --plugin-dir ./claude-learn

Two Layers of Learning

Layer 1 — Project Profile (always-on). Run /learn once in a project. The plugin detects your stack, profiles your current knowledge level per language/framework, and from that point on every coding conversation in the project is Socratic. Claude calibrates its guidance to your level and tracks the concepts you struggle with. Progress is saved across sessions.

Layer 2 — Topic Lessons (on-demand). /learn <topic> starts a structured, step-by-step lesson plan. Each step runs a two-phase cycle that repeats all the way through the plan:

1. Mini Lesson — Claude teaches the concept directly: a short, warm explanation with an everyday analogy, before any question is asked.
2. Problem Solving — Claude poses a Socratic question that makes you apply what you just learned. You can't advance to the next step without demonstrating genuine understanding via /check.

Then it repeats: Mini Lesson for the next step, Problem Solving for the next step, and so on. Claude will also offer to scaffold an exercises/ folder with any starter files or data the lesson needs. Progress (including which phase you're in) is saved across sessions.

Both layers track what you know per topic on a 1–10 scale and steer you toward your weakest areas.

Getting Started

# In any project directory:
/learn

Claude will detect your stack (Java, Python, React, etc.), ask a few quick questions to gauge your level, and write a knowledge profile. From then on, it's always in tutoring mode for that project.

# Start a lesson on a specific topic:
/learn dependency injection beginner
/learn react hooks
/learn binary search trees advanced

To set difficulty, include the word beginner, intermediate, or advanced anywhere in the command (it's pulled out as a whole word, so it won't get confused with similar-looking topic words). If you leave it out, Claude infers it from your profile level for that language/framework, or asks you directly rather than silently guessing intermediate.

Commands

Command	Description
/learn	Initialize project profile — detects stack, sets knowledge levels
/learn <topic> [difficulty]	Start a guided lesson. Difficulty: beginner, intermediate, advanced — inferred from your profile or asked if omitted
/hint	Get the next hint (1st time: directional, 2nd time: structural, 3rd time: near-explicit or partial solution). Only applies once you're in the Problem Solving phase
/check	Submit your understanding for evaluation — gates step progression. Only applies once you're in the Problem Solving phase
/progress	See your knowledge profile, active lesson status, and reinforcement queue
/explain <concept>	Get a concept explained without solving your current step
/reinforce	Targeted session on your highest-priority weak concept

What Claude Will and Won't Do

Will do:

• Teach each new concept directly in a short Mini Lesson before asking you to apply it
• Ask questions that guide your thinking (Socratic method) during Problem Solving — one small question at a time, in a warm, CS50-style teaching voice
• Calibrate depth to your knowledge level (beginner gets everyday analogies and plain language; advanced gets trade-offs and edge cases)
• Give progressively more specific hints — never the full answer
• Evaluate whether you understand (not just whether output is correct)
• Track weak concepts and surface them for reinforcement
• Resume lessons and profile across sessions