A Chinese-language book-style repo analyzing Claude Code architecture for Agent Harness. It provides structured chapter breakdowns, data points, and references, with no code samples or install instructions.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
The repository presents a book-length architectural analysis of Claude Code, titled "御舆：解码 Agent Harness". It describes the Agent Harness as the runtime framework that drives agent dialogue loops, tool permissions, and context management. The work emphasizes architecture analysis, design philosophy, and transferable cognitive models for different agent frameworks.
How it works
The book is organized into four parts:
- Part 1: Foundation — concepts like dialogue loops, tool systems, and permission pipelines.
- Part 2: Core systems — configuration, memory, context, and hooks.
- Part 3: Advanced patterns — agent composition, multi-agent orchestration, skills/plugins, MCP integration.
- Part 4: Engineering practice — streaming architecture, Plan mode, and building your own Harness. Each chapter focuses on design decisions, motivations, trade-offs, and preventing anti-patterns rather than API usage or prompts.
Contents include modules and chapters referencing internal components such as Tool interfaces, four-stage permission pipelines, memory models, context compression stages, hook lifecycles, sub-agent forks, coordination patterns, and external protocol integration. There are Appendix materials with maps, tool inventories, flag sheets, and glossaries.
Getting started
The repository README presents an English/Chinese bilingual presentation and a navigation path, but there are no installation commands, setup steps, or runnable code. The material appears to be a reference resource rather than a software library or toolkit. Direct quotes from the README include the sections describing goals, structure, and data points, but no setup instructions.
Recent releases
RELEASES (latest 0): - none
Traction
Stars: 4050; Forks: 805; Open issues: 15. The data shows star count but no recent release activity.
Behind the repo
The project cites Claude Code as the subject of its analysis and mentions Anthropic in a declaration, stating that the book analyzes public documentation and product behavior related to Claude Code.
Caveats
License is listed as license CC BY-NC-SA 4.0 in badges, but the README indicates no explicit code license in the repo description. Created 2026-03-31; last push 2026-06-19. The content is a design/architectural analysis with no runnable components or installation steps.