This repository presents a technical book, "Harness Engineering," focused on extracting AI coding agent architectural patterns, context strategies, permission systems, and production practices from the publicly released code of Claude Code v2.1.88. It targets engineers building AI coding platforms and developers seeking deep insights into the engineering implementation of advanced AI agents, offering practical, reverse-engineered knowledge.

How It Works

The book analyzes the publicly released package and source map de-compilation of Claude Code v2.1.88. It eschews official documentation replication, instead deriving architectural patterns, Agent Loop mechanics, tool execution orchestration, prompt engineering techniques (system, tool, and model-specific tuning), context management strategies (compression, token budgeting, Prompt Cache), and security/permission models directly from the codebase. This approach offers a practical, implementation-driven understanding of AI coding agent design.

Quick Start & Requirements

To build the book locally, use mdbook build book. To serve it, use mdbook serve book. The primary requirement is the mdbook tool.

• Chinese version: https://zhanghandong.github.io/harness-engineering-from-cc-to-ai-coding/
• English version (preview): https://zhanghandong.github.io/harness-engineering-from-cc-to-ai-coding/en/

Highlighted Details

• Focuses on reverse-engineering Claude Code v2.1.88's public release package and source map.
• Extracts AI coding Agent architecture patterns, context strategies, permission systems, and production practices.
• Covers Agent Loop, tool execution orchestration, prompt engineering, context management, Prompt Cache, security, multi-agent orchestration, and production best practices.
• Structured into 7 main parts and 4 appendices, detailing from underlying architecture to methodologies.

Maintenance & Community

The repository primarily tracks files necessary for deploying the book to GitHub Pages. Specific details regarding active maintenance, notable contributors, or community channels (like Discord/Slack) are not provided in the README.

Licensing & Compatibility

The README does not explicitly state a license for the book content or the repository. It clarifies that the content is for technical research and engineering discussion and does not represent Anthropic's official stance. Compatibility for commercial use or closed-source linking is undetermined without a specified license.

Limitations & Caveats

The analysis is based on reverse-engineering a public release package and source map, not official documentation. The English version is currently a preview with placeholder content, intended for future chapter-by-chapter translation. The repository's scope is limited to book deployment files.