Socratic 대화를 통해 가르치는 Claude Code 기술, 강의가 아닙니다. AI와 함께 작동하고 그들이 내장 된 것을 이해하려는 사람을위한 내장 - 그 코드, 구성, 워크플로우 또는 구조화 된 데이터 여부.

문제

모두 vibe 코딩입니다. AI는 코드를 작성, 당신은 앱을 발송. 그러나 무언가가 끊을 때, 당신은 붙어 있습니다. API 경로가 무엇인지 알 수 없습니다, 왜 await] 문제, 또는 당신의 fetch 호출이 실패하면 무슨 일이 발생.

그리고 그것은 단지 코드. PMs use AI to generate configs they don’t understand. Designers 프로토 타입과 프레임 워크를 디버깅할 수 없습니다. Researchers는 블랙 박스의 자동화 파이프라인을 구축합니다. 본은 동일합니다: AI는 일, 당신은 통제를 잃습니다.

이 기술이 그 문제를 해결합니다. 그것은 당신의 자신의 일에 대한 질문에 대한 질문을 통해 기초를 가르칩니다. - lecturing하지.

무엇 그것

• Socratic 메서드: 먼저 설명하지 마십시오. 당신이 생각하는 것을, 당신의 대답에 구축, 당신의 호기심을 따르십시오
• ** 5 레벨 힌트 사다리 ** : 열려있는 질문과 함께 시작, 만 마지막 리조트로 답변을 제공합니다 - 항상 후속 질문에 종료

/learn — Socratic Tutor for Vibe Coders

A Claude Code skill that teaches through Socratic dialogue, not lectures. Built for anyone who works with AI and wants to understand what they built — whether that’s code, configurations, workflows, or structured data.

The Problem

Everyone is vibe coding. AI writes the code, you ship the app. But when something breaks, you’re stuck. You don’t know what an API route does, why await matters, or what happens when your fetch call fails.

And it’s not just code. PMs use AI to generate configs they don’t understand. Designers prototype with frameworks they can’t debug. Researchers build automation pipelines that are black boxes. The pattern is the same: AI does the work, you lose control.

This skill fixes that. It teaches you the foundations by asking questions about your own work — not by lecturing.

What it does

• Socratic method: Never explains first. Asks what you think, builds on your answer, follows your curiosity
• 5-level hint ladder: Starts with open questions, only gives answers as a last resort — and always ends with a follow-up question
• Context cascade: Adapts to whatever you’re working with — git diffs, codebases, config files, markdown docs, or just a conversation. Never dead-ends.
• Adapts to you: Learns your background (design, PM, data, research) and tailors analogies to your world
• Frustration-aware: Detects when you’re stuck and changes approach instead of pushing harder
• Teach mode: When you’re confident, flip the roles — you teach, the skill plays dumb and asks naive questions
• Progress tracking — two files: a JSON brain for the skill (confidence, spaced repetition, open threads) AND a markdown journal in your voice (first-person, organized by thematic eras). The JSON tells the skill what to teach next; the journal tells you where you came from.
• Delight: Celebrates real growth moments, remembers your journey across sessions

Install

git clone https://github.com/deniz-miro/claude-learn-skill.git
cp -r claude-learn-skill/learn ~/.claude/skills/learn

That’s it. The skill is now available as /learn in any Claude Code session.

Where your progress lives

By default, ~/.claude/learning-progress.json. One file, persists across every project and every session on your machine. Switching repos, cloning a fresh checkout, or migrating a codebase doesn’t reset your progress — because the file lives next to your Claude config, not inside any single project.

Override: set $LEARN_PROGRESS_PATH to point elsewhere if you want a custom location (e.g. a synced cloud folder, or a sandbox-wide path that spans multiple project directories).