shareai-lab/learn-claude-code
Learn Claude Code is an open educational project that demonstrates how to construct a minimal coding agent harness from scratch. Structured across seventeen progressive lessons, the curriculum guides builders from foundational agent loops and shell-tool execution to advanced topics such as context compaction, worktree task isolation, and checkable goal loops. An included web simulator offers visual inspection of harness architectures.
The repository is aimed at software engineers and artificial intelligence researchers seeking to understand the mechanics of agent execution environments rather than relying on complex orchestration frameworks. The GitHub snapshot shows the repository is not archived and records an MIT license. The documentation highlights related alternative implementations from the same creators, including claw0 for always-on persistent agents, the Kode CLI package, and the standalone Kode Agent SDK.
Prerequisites include Git, Python, pip, an Anthropic API key, and Node.js with npm for the web viewer. Setup begins by running git clone https://github.com/shareAI-lab/learn-claude-code, followed by cd learn-claude-code, pip install -r requirements.txt, and copying .env.example to .env to set the key. Learners can test lessons directly using python s01_agent_loop/code.py, or launch the browser interface with cd web && npm install && npm run dev.
Data as of : 79784 stars · 12765 forks.