Claude Code Source MapAI coding agent architecture, reconstructed from public package artifacts
An unofficial source-map research project and bilingual tutorial for understanding local agent runtimes.
This site turns the public @anthropic-ai/claude-code package source-map reconstruction into a readable architecture guide. It is for research and education, not a claim about Anthropic's private repository or unreleased implementation.
AI coding agents are moving toward local runtimes that can search repositories, read and edit files, run commands behind permission checks, stream multi-turn work, and compact context during long sessions. This tutorial uses Claude Code as a concrete case study for that architecture.