Learn Claude Code

Active
GitHub Python MIT

Description

A nano claude code-like agent harness built from scratch, demonstrating how to build AI coding assistants from zero to one.

Key Features

  • Agent Harness architecture teaching - Systematic coverage of Tools + Knowledge + Observation + Action + Permissions five-layer architecture
  • Build Claude Code from scratch - Implement a complete AI coding assistant in Bash, covering file ops, shell execution, and context management
  • Harness engineering methodology - Deep exploration of the core philosophy: agency comes from training, harness comes from engineering
  • Tool implementation patterns - Teaching atomic, composable tool interface design (file I/O, shell, API calls, browser control)
  • Context management techniques - Sub-agent isolation, context compaction, and task persistence to prevent context bloat

Use Cases

💡 Understand how AI coding assistants work internally, learning the full pipeline from model to engineering
💡 Build agent harnesses for your own domain by migrating design patterns beyond coding scenarios
💡 Learn tool interface design, context permission management, and trajectory data collection for model fine-tuning
💡 Train team members as AI agent engineers with a shared harness engineering mental model

Tags

coding agent typescript framework cli

Categories

💻 Coding Agent

Quick Start

# Clone the repository
git clone https://github.com/shareAI-lab/learn-claude-code.git
cd learn-claude-code

# Read the tutorial docs and start building your Agent Harness from scratch
# Core concept: Harness = Tools + Knowledge + Observation + Action + Permissions
cat README.md
Visit GitHub Visit Website

Related Projects