Learn Claude Code

Learn Claude Code is an open-source educational repository that teaches developers how to build agent harnesses — the tools, context management, permissions, and environment infrastructure that surround a model and make it effective in a specific domain. Structured as 12 progressive sessions, it reverse-engineers Claude Code's architecture from a minimal agent loop all the way to isolated, autonomous multi-agent teams. The project is written primarily in TypeScript and Python, licensed under MIT, and includes an interactive Next.js web platform for visualizations and step-through diagrams.

• s01–s02: The Agent Loop & Tool Use — Start with python agents/s01_agent_loop.py to build the minimal while loop that drives every agent; add tools by registering handlers in a dispatch map.
• s03–s06: Planning & Knowledge — Learn TodoWrite for structured planning, subagent isolation with independent messages[], on-demand skill loading via tool_result, and three-layer context compression for infinite sessions.
• s07–s08: Persistence — Build a file-based task graph with dependency ordering; run slow operations in daemon threads with notification injection so the agent keeps thinking.
• s09–s12: Teams & Autonomy — Spawn persistent teammate agents with async JSONL mailboxes, implement request-response team protocols, enable autonomous task claiming, and isolate parallel execution in separate worktrees.
• Interactive Web Platform — Run cd web && npm install && npm run dev to access step-through diagrams, source viewer, and session documentation at localhost:3000.
• Multilingual Documentation — Mental-model-first docs (problem → solution → ASCII diagram → minimal code) available in English, Chinese, and Japanese.
• Companion CLI & SDK — After completing the sessions, apply knowledge via the open-source Kode Agent CLI (npm i -g @shareai-lab/kode) or the embeddable Kode Agent SDK for backend and browser-extension integration.
• Universal Harness Patterns — Design principles generalize beyond coding agents to farm management, hotel operations, medical research, manufacturing, and any domain requiring perception, reasoning, and action.