How AI Coding Agents Work: A Deep Technical Guide Through claw-code: Rust, Python, MCP, and the Architecture Behind the Leak

When Sigrid Jin's claw-code repository hit 50,000 GitHub stars in two hours, a record, it signalled something larger than a single project: a new era of AI-assisted software engineering, and a community demanding transparency in the AI tooling they rely on every day.claw-code is a clean-room Python and Rust reimplementation of the Claude Code AI coding agent harness. Built in a single overnight session using AI-orchestrated tooling, it mirrors the architecture of one of the most capable coding agents ever built, without reproducing any proprietary code.This book is the definitive technical guide to claw-code. Across 11 detailed chapters, you will:• Understand the architectural decisions behind a dual-language (Rust 72.9% / Python 27.1%) production-grade agent framework• Explore the six Rust crates that power the runtime: API streaming, CLI, tools, compat harness, commands, and the core ConversationRuntime generic• Master the Python orchestration layer: QueryEnginePort, PortRuntime, ToolPermissionContext, and the 27-subcommand CLI• Learn how the Model Context Protocol (MCP) works across 6 transport types, and how to extend claw-code with custom plugins, skills, and MCP servers• Study the parity audit system, a built-in mechanism for tracking feature completeness against any reference implementation• Understand the permission model, security architecture, and why matters for production AI agents• Apply advanced patterns: multi-agent swarm orchestration, SSE streaming, LSP integration, and building on top of claw-code for custom automation pipelines• Explore the Claude Code skills ecosystem, 49,000+ community stars, 1,000+ skills, cross-agent portability, and how to build your own SKILL.md extensionsWhether you are an AI/ML engineer building custom coding agents, a Rust developer looking for a real-world dual-language architecture to study, an open-source contributor wanting to participate in claw-code development, or a security researcher at the intersection of AI and software supply chains, this book gives you the complete picture.Every chapter includes practical code examples, architecture diagrams, and immediately actionable recipes. The book is grounded entirely in publicly available open-source code and public security research, nothing proprietary, nothing speculative.The future of AI-assisted development is open. This book shows you how it works. Read more