Claude Code Source MapAI coding agent architecture research and tutorial

Based on Claude Code v2.1.88 public npm package and source-map analysis. A bilingual course on local agent runtime architecture.

Start Learning

Research Note

GitHub Repo

🔄

Agent Loop

Follow QueryEngine, tool calls, tool results, and follow-up model turns to see why a coding agent is an execution loop, not one answer.

🛠️

Tool System

8 core tools let AI read files, write code, and run commands. Learn how they're defined, registered, and invoked.

🛡️

Security Design

A five-layer permission model makes AI both powerful and controllable. From tool-level checks to user hooks.

🚀

Architecture Course

The value is not cloning a product. It is turning public package analysis and source-map reconstruction into a learnable AI coding agent architecture path.

Research Positioning

This project is unofficial research material, not an Anthropic source release. Its architectural readings are based on public npm package artifacts, source-map reconstruction, and analysis already present in this repository.

Market Context: From Chat Plugin to Local Agent Runtime

AI coding agents are moving from "chat plugin inside an editor" toward "local runtime near the repository and terminal." A useful coding agent needs file search, code edits, command execution, permission checks, streaming multi-turn progress, and context compaction when sessions get long.

This repository's original value is turning a complex source-map reconstruction into a learnable architecture course: tool loop, permission model, grep/file context, streaming multi-turn architecture, and context compaction.