The skill behind this guide: AI Python Tutor. Turns Claude, ChatGPT or Gemini into a tutor that makes you write the Python yourself - $19, one payment, yours permanently.
View the Python tutor →Used as a Python tutor, Claude is good at one narrow thing and weak at the thing most people expect. It cannot see your editor, it does not know what your screen says, and a tutor worth the name will not run the exercise for you even where the tool could. What it does well is sit on the other side of a task: set the exercise, read the code you paste back, point at the line where your idea and the interpreter disagree, and make you fix it yourself. That is tutoring. The typing stays yours.
That distinction is the whole reason this works. Most people arrive after a course, not before one: forty hours of watching somebody else write Python, and still no way to start from an empty file. A chat window that hands over finished code makes that worse. One running a tutor file makes you produce the code and intervenes only when you are stuck, which is the part a video cannot do.
What does an AI Python tutor actually do?
Four things, in order. It finds your real level with a task rather than a quiz, because people describe their own level badly in both directions. It builds a lesson arc from there instead of starting at variables for the fourth time. It gives you one explanation, one worked example and one task. Then it reads what you wrote and tells you where it goes wrong, not what to type instead.
Be clear about the mechanics. Claude reads text and images, so it handles pasted code and a screenshot of a traceback perfectly well. It is not attached to your machine: it does not watch you type, it does not know your Python version unless you say, and it cannot see whether you ran the thing. Claude can execute code in some settings, but a tutor that runs your exercise for you has taught you nothing, so the AI Python Tutor file is written the other way round: you run it, you paste what happened, it works backwards from there.
Why does the course you finished not stick?
Because reading working code and writing code from nothing are different skills, and only one of them gets practised in a video. Following along feels like learning: you recognise every line as it appears, nothing surprises you, and the file at the end runs. Then the instructor stops and the blank file is as blank as it was in week one. What closes that gap is producing wrong code and being made to repair it, which is the loop the tutor file is built around: short task, your attempt, a diagnosis, a second attempt.
How do you set Claude up as a Python tutor?
Whether you are running the skill file or starting from scratch, the first move is the same: make it place you before it teaches you. Paste this into a new conversation and do the task honestly, including the part where you cannot finish it.
You are my Python tutor. Do not teach anything yet.
Give me one small task, not a quiz, that will show you what I can actually do. It should be solvable in ten to fifteen lines and it should touch a loop, a list and a function.
I will write it myself and paste it back, working or not. Then tell me three things:
1. What my code shows I already understand.
2. The first real gap, the one that will block everything after it.
3. The order you would teach me the next six topics, and why that order and not another.
Do not write the task solution for me at any point. If I paste something broken, that is the useful case, work with it.
What does one lesson look like?
One concept, one example short enough to read in full, one task, one check. Anything longer and you are back to watching. The arc that works for most people is lists and loops, then functions, then dictionaries, then files, then errors and tracebacks, then classes, and only then the libraries people came for, usually requests and pandas. Skipping to pandas is the most common way to waste a month.
Teach me [TOPIC] as a single lesson in four parts, in this order:
1. EXPLANATION - under 150 words, plain language, no analogy about restaurants or boxes.
2. EXAMPLE - one snippet under 12 lines. Walk through what happens line by line, including what each variable holds.
3. MY TASK - a problem I write from scratch that cannot be solved by editing your example. Tell me what correct output looks like so I can check myself.
4. WAIT. Do not continue until I paste my attempt.
When I paste it: tell me whether it works, what it would print, and where my reasoning is off. Give me a corrected version only if I have failed the same task twice and ask for it.
The Python tutor teaches a subject over weeks. This one handles the other situation: an assignment in front of you tonight, in any subject, worked through as steps you can defend in class rather than an answer you cannot explain. $29, one payment, and it works in Claude, ChatGPT and Gemini like everything else here.
View the Homework Helper →How do you get it to debug with you instead of for you?
This is where most people accidentally destroy the value of the whole thing. You paste a traceback, get a corrected file back, paste that into your editor, it runs, and you have learned nothing except that pasting works. Forbid the answer in the prompt.
Here is my code and the full error it produced.
[PASTE CODE]
[PASTE TRACEBACK]
Do not give me corrected code. Instead:
1. Read the traceback out loud to me. Which line does Python actually blame, and which line is really at fault? Those are often different.
2. Tell me what that error type means in general, in one sentence.
3. Ask me one question about what I expected a specific variable to contain at that line.
Then stop and wait for my answer. If my answer is wrong, ask a second question rather than telling me.
Once it is fixed, tell me the one habit that would have stopped this class of bug happening again.
Reading a traceback is a large part of learning Python and the part courses skip. The same prompt works for SQL errors if you are running the AI SQL Tutor alongside it, which people learning Python for data work usually end up doing.
Where does it fail?
- It cannot see your environment. Virtual environments, PATH problems, the wrong interpreter selected in your editor: the real first-week blockers, and it has no way to look.
- It cannot confirm your code ran. Say the output was correct and it believes you. Progress depends on you being honest about what actually happened.
- It slides back into answering. Ask twice in a row for a hint and it starts giving solutions again. Re-paste the rule when you notice it.
- Library details go stale. An API that changed since its training data gets recalled with full confidence. Check anything library-specific against the official docs.
- It is not a substitute for building something. The lessons get you to competence. A project you want to exist is what fixes it in place.
- It has no memory across chats unless you give it one. Keep the tutor in one conversation or one Project, otherwise every session restarts.
A tutor deliberately slows you down. Once you can write Python unaided you want the opposite: a working assistant for structure, review and the parts of a real codebase nobody teaches in a course. Ambrose is that file, $29, one payment. Learn with the tutor first, then switch.
View Ambrose →How do you install the file?
The tutor is a SKILL.md file inside a ZIP. In Claude, open Customize → Skills, press the + button, choose Create skill, then Upload a skill, and select the ZIP. Toggle it on and it is live. Skills are available on the Free, Pro, Max, Team and Enterprise plans.
The one real requirement is code execution. If the Skills menu is greyed out, turn it on under Settings → Capabilities, where it is listed as code execution and file creation. On Team and Enterprise plans an owner controls it in Organization settings, so if it is switched off org-wide, that is who to ask. Anthropic documents all of this in its Skills help article.
If you cannot enable code execution, or you are working in ChatGPT or Gemini, the fallback uses the same file: open the ZIP, copy the contents of SKILL.md, and paste it into a Project, a custom GPT, a Gem or plain custom instructions. The behaviour is identical, you just re-paste it when you move tools.
Learning something other than Python?
The tutor files are one family with one method, so the language changes and the loop does not. If the browser is where you want to be, the AI JavaScript Tutor covers the same ground with the DOM and async in place of files and classes. For a first typed language there is the AI Java Tutor, and for the Microsoft side the AI C# Tutor. Once Python is in hand, the AI Django Tutor is the usual next step if you want a web application at the end, and the AI SQL Tutor is the one people wish they had started earlier. The full learning and research collection has the rest.
In summary:
Use the AI Python Tutor ($19) to be placed properly, taught in one-concept lessons, and made to write every line yourself. Keep AI Homework Helper ($29) for the assignment that is due tomorrow, and move to Ambrose ($29) once you no longer need teaching. All work in Claude, ChatGPT and Gemini, no coding required to install, 30-day money-back guarantee.
AI Python Tutor
One file, pasted once. It places you with a task, builds the lesson arc from where you actually are, and refuses to write your code for you. No subscription.
KissMySkills is a marketplace of 853 AI skills, 158 prompt packs, 55 agents & free tools for Claude, ChatGPT & any AI chat.
Related Skill Guides
- How to Use Claude as a JavaScript Tutor (In the Console, Not in a Video)
- How to Use Claude as an Excel Tutor (On Your Own Messy Data)
- How to Use Claude as a Homework Helper and Study Buddy
- How to Use Claude for Backend Development: The Niko Skill Guide
- Claude Skill Examples: 12 Real SKILL.md Files Explained
- Best Claude Skills in 2026