Mini Claw Code

Build a coding agent from scratch in Rust — guided by Claude Code's architecture.

Read the Book (EN) · 中文版 (Book) · 中文 README · Quick Start · Chapters

---

You use coding agents every day. Ever wonder how they actually work?

It's simpler than you think. Strip away the UI, the streaming, the model routing — and every coding agent is just this loop:

loop:
    response = llm(messages, tools)
    if response.done:
        break
    for call in response.tool_calls:
        result = execute(call)
        messages.append(result)

The LLM never touches your filesystem. It *asks* your code to run tools — read a file, execute a command, edit code — and your code *does*. That loop is the entire idea.

This book builds that loop from scratch and then grows it into the full architecture of a real coding agent: streaming, permissions, hooks, plan mode, configuration, and more. **18 chapters. Test-driven. No magic.**

What you'll build

A working coding agent that can:

• Run shell commands — ls, grep, git, anything
• Read, write, and edit files — full filesystem access with surgical find-and-replace
• Talk to real LLMs — via OpenRouter (free tier available, no credit card)
• Stream responses — SSE parsing, token-by-token output
• Search a codebase — glob for files, grep f