// the find
24kchengYe/human-skill-tree
🌳 AI-Powered Skill Tree for Lifelong Human Learning. 30+ skills from K-12 to career & social intelligence, built on cognitive science. | 人类养成记：AI 驱动的终身学习技能树
A collection of 33 markdown-based prompt files ("skills") that instruct AI assistants to teach using Socratic methods, spaced repetition, and active recall instead of just answering questions. There's also a Next.js web app (the more substantial piece) that wraps these skills into a multi-tutor classroom experience. Aimed at students and self-learners who want AI to teach rather than just answer.
The Socratic framing is actually well-thought-out — the debug example (guide the student to find the bug vs. just fixing it) is a concrete, useful pattern that most AI interactions skip entirely. The web app's multi-agent classroom (teacher + assistant + AI students discussing together) is a genuinely novel UI idea for learning, not just a chat window. The learning science citations are real and relevant — this isn't cargo-culting buzzwords, the cited PNAS studies actually support the design choices. The skill file structure is simple enough that adding a new subject is just writing a markdown file, which makes community contributions tractable.
The "skills" are just prompt files — there's no actual spaced repetition algorithm, no interval scheduling, no persistent learner model. You get the vocabulary of spaced repetition without the mechanism; real SR requires tracking what you got wrong and when to show it again, none of which this does. The web app depends on OpenRouter with 18 models, but the actual pedagogical behavior is entirely prompt-dependent — there's no guarantee the 7-layer system prompt survives context compression in a long session, meaning the Socratic guardrails can quietly fall apart mid-conversation. Calling this "800+ subjects" is inflated; it's 33 skill categories, each of which claims broad coverage via a single prompt file — the depth per subject is whatever the LLM brings, not curated content. The AGPL-3.0 license on the app is a meaningful adoption blocker for anyone wanting to build something commercial on top of this.