your-ai-tutor docs

your-ai-tutor (yat) runs inside Claude Code. You open claude in this repo and say what you want to learn. From then on the tutor asks, and you answer, predict and explain back. It never finishes the work for you.

Everything is plain files. Each topic you study is an env, a folder under envs/. Each thing the tutor knows how to do is a skill, triggered by a phrase you say. What the tutor learns about you lives in memory, and every line of it points back to something you did.

Find your starting point

You want to	Read first
Start learning a topic today	Getting started
Know what to say to the tutor	Everything you can say
Understand why it won't just give you the answer	Concept
Run a learning session	Learn a topic
Know what to do today, or pick up where you stopped	Sessions and today's block
Quiz yourself on what you've covered	Quiz, test and assignments
Read a book or paper with the tutor	Read along
See what you know, and why the tutor thinks so	Mastery map
Learn how a website works by taking it apart	Reverse engineering
Change how the tutor behaves	Architecture
Write a new skill	Skill API

One topic, start to finish

Here is what a run looks like, using linear algebra as the example topic. Each step is one phrase.

1. I want to learn linear algebra. The tutor asks five short questions, one at a time, and creates envs/linear-algebra/ with a spec, a roadmap and a first bookmark. It ends by telling you the phrase to start.
2. let's do linear-algebra. A session starts. Before anything is shown you predict what will happen. When you're stuck, the tutor climbs a hint ladder one rung at a time. A unit only counts once you've explained it back in your own words.
3. card this. Optional: a flashcard in Anki, in your wording.
4. wrap up. Your spot is saved, the session goes into the env's journal, and today's box is ticked.
5. what should I do today. Next time, the tutor names today's block first, then opens with a question built from the last thing you missed. No recap.
6. quiz me. Questions only on what you've covered, never repeated. For each miss you write one correction line in your own words.
7. build my mastery map. You name the ideas behind your roadmap. Each idea's state moves only when something you did is on record.
8. retro. A debrief turns what you noticed into proposals for changing the env or the harness. A person reviews them in git before anything changes.

Turn on only what you need

The core needs nothing but Claude Code. Each extra stays off until you ask for it, and the skill tells you what to set up the first time.

• Flashcards need Anki desktop with the AnkiConnect add-on. See Anki cards.
• Reverse engineering a live site needs the Claude-in-Chrome extension. See Reverse engineering.
• Study material from PDFs or video needs pdftotext and ffmpeg. See Study material.

Your data stays yours

envs/, media/, tracker/ and memory/ ship empty and are git-ignored, so your progress never lands in a public fork by accident. Remove the lines at the bottom of .gitignore if you want to version them in a private repo.