Claude Code v2.1.88

The Source Decoded

A comprehensive technical deep-dive into the architecture, tools, and internals of @anthropic-ai/claude-code v2.1.88 — reconstructed from public source maps.

1,884

TypeScript Files

39

Tools

101

Commands

512,000+

Lines of Code

36

Service Modules

17

Built-in Skills

Tool System

39 specialized tools from BashTool to AgentTool, each with its own permission model

Security Model

Three-layer permission system: tool-level rules, file-level isolation, sandbox mode

Multi-Agent

Coordinator mode enabling Agent swarms, team management, and message passing

MCP Protocol

Full Model Context Protocol support across Stdio, SSE, HTTP, and WebSocket transports

IDE Bridge

Bidirectional communication bridge for VS Code and JetBrains extensions

Performance

Parallel Keychain prefetch, feature-flag dead code elimination, lazy loading

4-Layer Architecture

Entry Layer

CLI parsing, bootstrap, env initialization

Query Engine

LLM orchestration, streaming, tool loop

Tool & Service Layer

39 tools + 36 service modules

UI Layer

React/Ink terminal UI, 111 components

Bootstrap Flow

1.main.tsx entry
2.Parallel: MDM prefetch + Keychain prefetch
3.Parse CLI args (Commander.js)
4.Load config, OAuth, policy limits
5.Initialize MCP connections
6.Launch React/Ink UI + QueryEngine