learn-claude-code

Why we included this project

Most agent frameworks ship as finished products and expect you to trust what happens inside. This one takes the opposite route: it walks through building a Claude Code style agent harness from scratch, starting from the idea that agency lives in the model while the surrounding code is just the vehicle. For developers who want to see what actually happens between a model call and a completed task, the step-by-step build shows the moving parts, including tool calling, context management, and the shell loop that ties them together. It reads more like a structured course than a library, so treat it as a learning resource to work through rather than a dependency to drop into production. If you are designing your own agent product, the real takeaway is the reasoning behind the design choices, not the code itself.