Learn Claude Code

Why it's ranked here

The project earns attention because it exposes concrete mechanics instead of stopping at agent terminology. Its strongest material shows how context isolation, dependency tracking, and asynchronous communication fit around a stable core loop. The opinionated framing is useful, though broader than the supplied implementations prove.

What's good

Lessons isolate one change at a time and explain the problem before showing the mechanism. Path checks demonstrate workspace containment. Subagents discard noisy histories and return summaries. Disk-backed dependency graphs survive context compression. Team examples make lifecycle states and mailbox communication tangible rather than abstract.

Tradeoffs

Two tutorial tracks coexist, and their chapter numbers do not always align. Readers must stay within one track. Several examples use deliberately simple controls: fixed iteration limits, threaded workers, file-based coordination, and mailboxes that empty when read. These choices teach clearly but leave reliability questions beyond the demonstrated scope.

How to use it well

Use it as a hands-on architecture course for developers building model-driven coding tools or similar operational environments. Follow the current lessons in order, run each example, and compare every added mechanism with the unchanged loop. It does not cover training models, and it should not substitute for production hardening guidance.

Observed

Primary language

Python

Install surface

Python dependencies are declared in requirements.txt for installation with standard Python package tooling.

Model interface

Uses the Anthropic Python SDK, with a declared minimum dependency of 0.25.0.

User interface

Runnable command-line lesson programs accept prompts and expose commands for team and inbox inspection.

Documentation languages

English, Chinese, and Japanese documentation are provided.

Repository structure

A canonical 17-lesson track coexists with a preserved legacy 12-lesson track.