This guide helps you set up Bloom One-vs-One Study from scratch and start your first 1-on-1 AI tutoring session.
You need:
- Claude Code (Anthropic's official CLI tool)
- A terminal (macOS Terminal / iTerm2 / Windows Terminal / whatever you prefer)
- A text editor (VS Code, Cursor, etc., for reading and annotating documents)
If you haven't installed Claude Code yet:
npm install -g @anthropic-ai/claude-codeAfter installation, run claude to confirm it launches properly. First run requires logging in to your Anthropic account.
If you're unsure what Claude Code is: it's a CLI tool that lets you chat with Claude in your terminal, and Claude can directly read and write your local files. This is the foundation of how this system works.
git clone https://github.com/Li-Evan/Bloom-one-vs-one-study.git
cd Bloom-one-vs-one-studyInstall the bundled tutor skill into this clone's local Claude Code skills directory:
mkdir -p .claude/skills
cp -R skills/bloom-tutor .claude/skills/This keeps the public tutoring protocol in skills/bloom-tutor/ instead of relying on local agent instruction files.
Start Claude Code in the repository directory:
claudeAsk Claude to use the bloom-tutor skill when starting or continuing a course.
In the Claude Code conversation, type:
Create a new folder and help me learn [your topic]
For example:
Create a new folder and help me learn Python decorators
Create a new folder and help me learn game theory basics
Create a new folder and help me learn personal income tax
Claude will immediately generate:
syllabus.md— course syllabus (defines all abilities you'll master)01.md— your first lesson document
Setup complete. Here's how to use it.
Open the generated .md file in your text editor. Each document contains:
- Prerequisites / Difficulty / Estimated reading time
- Main content (knowledge with bold annotations and examples)
- Thought questions (2–3 questions, no answers given, designed to deepen your thinking)
- Feedback section (where you write your feedback)
While reading, write the following anywhere you feel confused:
???[Why use recursion here instead of a loop?]
Or use full-width question marks:
???[What's the intuitive meaning of this formula?]
You can place annotations anywhere in the text, as many as you want. These annotations are the most authentic snapshot of your thinking, and the tutor prioritizes them.
At the bottom of the document in the "Your Feedback" section, write:
- Your answers to the thought questions (try to reason through them yourself — wrong answers are fine)
- Your insights, confusions, or topics you'd like the next lesson to dive deeper into
- Anything else you want to say
Go back to the Claude Code terminal and say:
I've finished reading
The tutor will:
- Read all your annotations and feedback
- Possibly ask you 1–2 key questions (max 2 rounds, no endless grilling)
- Generate the next document
The next document's opening will include:
- Thought question review (evaluates each of your answers, provides correct answers)
- ??? responses (addresses every confusion you annotated)
- New content (tailored to your understanding level)
When all mastery items in the syllabus are covered, the system automatically generates an evaluation article (no new content — final understanding confirmation). After reading the evaluation, the system auto-generates summary.md (a complete course summary).
During your learning, if you encounter a particularly important insight you want in the final summary, annotate it:
#summary:[The essence of option pricing is replication — constructing a portfolio of known-price assets that reproduces the same cash flows]
The #-less format also works:
summary:[This analogy is brilliant — Nash equilibrium in game theory is like a traffic jam — no one can benefit by unilaterally changing routes]
These materials are automatically collected and integrated into summary.md.
Topics can be organized by category:
Create a new folder under CFA, help me learn fixed income
This creates a CFA/fixed-income/ directory.
You can study multiple topics simultaneously. When entering Claude Code, tell the tutor which topic you'd like to continue:
I want to continue studying Python decorators, I've finished reading 02.md
| Command | Action |
|---|---|
/organize-learning |
Scan all topics, log new documents to the learning journal |
/view-learning-log |
View historical learning records (newest first) |
The system itself is completely free and open-source. You need Claude Code access (requires an Anthropic account).
Anything you want to learn — programming, finance, philosophy, psychology, math, history... no limits.
No. This is a core design principle. The essence of 1-on-1 tutoring is that every step adjusts based on your feedback. Batch generation would break this feedback loop.
Entirely on your local filesystem, in the cloned repository directory. No data is uploaded to the cloud. You can use Git to version-control your learning history.
The bundled skills/bloom-tutor package is designed for Claude Code Skills. Other AI agents can work if you import the same instructions into their equivalent skill/instruction system, but results may vary.
Anytime. Write your desired direction change in the feedback section, and the tutor will adapt in the next lesson. Mastery items in the syllabus are the goals; the path is entirely flexible.
This system is built on a simple belief:
The best learning isn't being lectured — it's being guided to discover.
Traditional online courses are one-directional — pre-recorded videos won't pause to explain your confusions. ChatGPT-style Q&A is fragmented — you get answers, but no system.
This system aims to balance both: systematic adaptive learning. The syllabus ensures you stay on track, and the feedback loop ensures content always matches your level.
Bloom proved that 1-on-1 tutoring achieves +2σ. We believe a well-designed AI agent can approach this effect.