AI Learning Workflow

This repository is a provider-agnostic wrapper around the ideas from amosblomqvist/learn, adapted so the same teaching workflow can be used with external agent environments such as Claude, Codex, and GitHub Copilot.

The original project is centered around a teaching system built around strong fundamentals, dependency graphs, and structured learning loops. This fork keeps that core approach but separates the reusable instructions from the provider-specific adapter files, making it easier to use the same agent design across different tools and platforms.

What this repo does

It provides:

• a shared, canonical set of agent instructions
• reusable teaching and visualization skills
• provider-specific adapter folders so the same behavior can be exposed through different agent runtimes
• a structure that is easy to extend with new roles, tools, and workflows

The goal is not just to let an AI answer questions, but to make it teach in a way that builds real understanding instead of memorized facts.

Why this exists

The base learn project demonstrates a powerful pattern: use a small number of strong principles, a clear teaching process, and explicit dependency-driven explanations.

This repository adapts that pattern for agent ecosystems outside the original environment by dividing the project into:

• shared prompts and rules in AGENTS.md, agents/, and skills/
• provider adapters in .github/, .claude/, and .agents/

That makes the same underlying teaching system portable across multiple AI coding assistants.

Repository structure

• AGENTS.md — top-level instructions for the project and shared behavioral rules
• agents/ — canonical agent definitions used as the base behavior
• skills/ — reusable skill files such as teaching and visualization workflows
• .github/ — GitHub Copilot adapter files
• .claude/ — Claude adapter files
• .agents/ — additional provider-specific adapter layer
• .gitignore — repo-local ignore rules

Core idea

The system follows a teaching loop built around:

• understanding the learner's current level
• finding the actual learning goal
• planning the dependency graph of concepts
• teaching from foundations upward
• checking understanding before moving forward

This is encoded in the teaching skill and the role prompts, and it is intentionally designed to work across different AI environments.

Example usage

This repository is meant to be used as a reusable prompt/agent library. The same conceptual teaching framework can be plugged into different environments:

• Claude Code via .claude/
• GitHub Copilot via .github/
• Codex or similar agent runtimes via provider-specific adapters or custom wrappers

In other words, the shared logic stays in one place, while the surrounding runtime integration remains lightweight and portable.

Design philosophy

This repo keeps the original learning system's philosophy:

• start from unconditional truths
• make the path of discovery explicit
• build understanding through dependencies
• test understanding before moving on
• prefer durable understanding over shallow recall

Notes

This is a configuration and prompt-oriented repository rather than an application server or package. It is best thought of as a reusable teaching scaffold for AI agents.

If you want to adapt it further, the easiest path is:

1. keep the canonical instructions in AGENTS.md and agents/
2. add or modify provider adapters in the runtime-specific folders
3. expand skills/ with new reusable capabilities as your workflow grows

Credits

Inspired by the teaching-oriented agent work in amosblomqvist/learn, with adaptations for multi-provider agent usage across Claude, Codex, and Copilot-style environments.