AI AGENTS
learn-claude-code
by shareAI-lab
Overview
A nano Claude Code-like agent harness project for learning harness engineering for real agents.
Details
learn-claude-code is a GitHub project from shareAI-lab described as “Bash is all you need” and “a nano claude code-like agent harness, built from 0 to 1.” Its README is described as teaching harness engineering for real agents, with a current 20-lesson track, setup commands using ANTHROPIC_API_KEY, and runnable Python lessons. The s01_agent_loop chapter documents the core loop: send messages and tools to the model, execute requested tool calls, feed results back, and continue until no tool is called.
When to Use
Use it to study how a small Claude Code-like agent harness can be built from the ground up. Use it when you want runnable lessons for agent harness engineering, including a basic agent loop with tool calls. Use it as a reference for understanding the workflow of sending messages/tools to a model, executing tool calls, and feeding results back.
Getting Started
- Open the GitHub repository at https://github.com/shareAI-lab/learn-claude-code.
- Read the README for the lesson track and setup commands using ANTHROPIC_API_KEY.
- Review the s01_agent_loop chapter to understand the documented core agent-loop workflow.
- Run or inspect the runnable Python lessons before adapting the harness ideas to your own project.
Key Features
- •Nano Claude Code-like agent harness built from 0 to 1.
- •README-described 20-lesson learning track for harness engineering.
- •Setup flow using ANTHROPIC_API_KEY.
- •Runnable Python lessons described in the README.
- •Documented agent loop that sends messages and tools, executes requested tool calls, and feeds results back.
Capabilities
- •agent-harness-learning
- •agent-loop-implementation
- •tool-call-workflow
- •python-lessons
Last updated May 29, 2026