claude-code-analysis

This repository is a structured static analysis of Claude Code's TypeScript source code, which became publicly accessible in March 2026 after Anthropic shipped npm packages with source maps intact, exposing 1,902 source files and 513,237 lines of code. Organized into eighteen chapters and multiple appendices, the project documents Claude Code's internal architecture across topics including the agent execution kernel (query.ts and QueryEngine.ts), multi-layer memory and session storage, sandbox isolation, tool call permissions, MCP integration, Skills extension mechanisms, prompt management, and multi-agent coordination. A dedicated section breaks down the TUI component tree under src/components/, and a competitive comparison chapter benchmarks Claude Code against Codex, Gemini CLI, Aider, and Cursor. The repository also catalogs hidden commands, feature flags, and a frustration-signal detection mechanism based on negative keyword analysis. Security researchers, AI engineers, and developers building on or competing with agentic coding tools are the primary audience. The project is presented strictly as academic research and carries no affiliation with Anthropic.

What people ask about claude-code-analysis