Model: claude-opus-4-6 (anthropic/claude-opus-4-6) Tools: OpenCode Generated: 2025年02月17日 GitHub: https://github.com/0xtresser/OpenCode-Book

中文版本

As of OpenCode version 1.2.5.

This book is divided into five parts and eighteen chapters, systematically covering OpenCode's architecture design and implementation principles — from foundational concepts to source code analysis, from core mechanisms to ecosystem extensions, and from theory to practice.

Guiding readers to understand the technical background and industry context behind OpenCode's creation.

• 1.1 From Auto-Completion to Agentic Coding
• 1.2 AI Programming Assistant Product Landscape
• 1.3 OpenCode's Positioning and Design Philosophy

Setting up the development environment and building a macro-level understanding of the project structure.

• 2.1 Obtaining Source Code and Environment Setup
• 2.2 Monorepo Structure Analysis
• 2.3 Core Package packages/opencode/src/ Module Map

Understanding OpenCode's architectural layers and data flow from a global perspective.

• 3.1 Client-Server Architecture
• 3.2 Core Data Flow: From User Input to AI Response
• 3.3 Multi-Project and Multi-Worktree Support
• 3.4 Layered Configuration System Design

The session is OpenCode's core abstraction — understanding it means understanding the entire conversation management mechanism.

• 4.1 Session Data Model
• 4.2 Message Model
• 4.3 SessionPrompt: The Session Entry Point
• 4.4 SessionProcessor: The Core of the Agentic Loop
• 4.5 Compaction: Context Window Management
• 4.6 Session State Management

Tools are the only way Agents interact with the external world.

• 5.1 Tool Abstraction Layer Design
• 5.2 ToolRegistry: Registration and Discovery
• 5.3 Built-in Tools Source Code Analysis
• 5.4 Permission Checks in Tool Invocations

An Agent is the abstraction that gives an LLM its "personality" and "capability boundaries."

• 6.1 Agent Data Model
• 6.2 Built-in Agent: build
• 6.3 Agent Auxiliary Functions
• 6.4 System Prompt Architecture
• 6.5 Custom Agent Configuration

Understanding how OpenCode uniformly interfaces with 20+ LLM Providers.

• 7.1 Vercel AI SDK Integration
• 7.2 Provider Registration and Resolution
• 7.3 Model Metadata System
• 7.4 ProviderTransform: Differentiated Model Adaptation
• 7.5 LLM Module: Streaming Call Implementation

MCP is the core protocol for OpenCode to connect with external tool ecosystems.

• 8.1 MCP Protocol Introduction
• 8.2 MCP Client Implementation
• 8.3 MCP OAuth Authentication
• 8.4 MCP Configuration

Security is the lifeline of AI Agents — deeply understanding how OpenCode controls Agent behavior boundaries.

• 9.1 Permission Model Design
• 9.2 Permission Evaluation Engine
• 9.3 Permission Configuration Hierarchy
• 9.4 Permission Request Interaction Flow
• 9.5 Permission Strategies for Each Tool

Understanding how OpenCode safely manages file changes and rollbacks.

• 10.1 Git Snapshot System
• 10.2 File System Tools
• 10.3 Patch System
• 10.4 Worktree Management

Understanding the core mechanism for decoupled inter-module communication.

• 11.1 Event Bus
• 11.2 Key Event List
• 11.3 Scheduler: Timed Task Scheduling

Understanding OpenCode's frontend implementation from a user interaction perspective.

• 12.1 CLI Startup Flow
• 12.2 TUI Implementation
• 12.3 Web UI
• 12.4 Non-Interactive Mode

Understanding the core of OpenCode's extensibility — the Plugin architecture.

• 13.1 Plugin Interface Definition
• 13.2 Plugin Lifecycle Hooks
• 13.3 Plugin Loading Mechanism
• 13.4 OpenCode SDK

Skills are modular extension units for Agent capabilities.

• 14.1 Skill Definition and Structure
• 14.2 Skill Loading Flow
• 14.3 Command System

As the most complex third-party Plugin in the OpenCode ecosystem, oh-my-opencode demonstrates the full power of the Plugin system.

• 15.1 Project Overview and Architecture
• 15.2 Plugin Entry Point and Initialization
• 15.3 Multi-Agent System
• 15.4 Custom Tool System
• 15.5 Hook System (53 Hooks Explained)
• 15.6 Feature Modules
• 15.7 Plugin Interface Implementation
• 15.8 Embedded MCP Services

Understanding how OpenCode integrates with various IDEs.

• 16.1 VSCode Extension
• 16.2 Zed Extension
• 16.3 ACP (Agent Client Protocol)

Through five progressive experiments, build key components of an AI programming assistant from scratch.

• 17.1 Lab 1: Implementing a Simple LLM CLI Chat Tool
• 17.2 Lab 2: Adding Tool Use Capability to the CLI Tool
• 17.3 Lab 3: Building an MCP Server
• 17.4 Lab 4: Writing an OpenCode Plugin
• 17.5 Lab 5: Building a Simplified Multi-Agent Orchestration System

Summarizing the entire book, distilling design wisdom, and looking ahead to future directions.

• 18.1 Review of OpenCode's Key Design Decisions
• 18.2 Design Insights from oh-my-opencode
• 18.3 Best Practices for Building AI Programming Assistants
• 18.4 Future Outlook

• Complete list and description of opencode.json configuration options

• Complete list and description of oh-my-opencode.json configuration options

• Parameters, return values, and permission requirements for each built-in tool

• Core concepts, transport methods, message formats

• Definitions of terms: LLM, Token, Context Window, Function Calling, Agent, MCP, ACP, LSP, AST, SSE, Monorepo, TUI, etc.

• Official documentation links
• Related papers and articles
• Recommended open-source projects

Quick reference table for key source code files covered in this book