claude-code-best

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

claude-code-best

Skill by ara.so — Claude Code Skills collection.

Claude Code Best (CCB) is a reverse-engineered, production-ready implementation of Anthropic's Claude Code CLI. It's written in TypeScript, runs on Bun or Node.js, and provides enterprise-grade features including multi-instance orchestration, ACP protocol support, Langfuse monitoring, remote control, web search, voice mode, and computer/browser automation.

由ara.so提供的技能 —— Claude Code技能合集。

Claude Code Best（CCB）是Anthropic旗下Claude Code CLI的一个逆向工程实现、可用于生产环境的版本。它使用TypeScript编写，可在Bun或Node.js上运行，并提供企业级功能，包括多实例编排、ACP协议支持、Langfuse监控、远程控制、网页搜索、语音模式以及电脑/浏览器自动化。

What It Does

功能介绍

CCB is an AI coding agent terminal interface that:

Runs Claude (or compatible models) in your terminal with full file system access
Supports multi-instance coordination via Pipe IPC and LAN discovery
Integrates with Zed, Cursor, and other IDEs via ACP protocol
Provides enterprise monitoring (Langfuse, Sentry, GrowthBook)
Offers web search, voice input, computer/browser control
Works with any Anthropic-compatible API provider (OpenAI, Gemini, etc.)

CCB是一款AI编码代理终端界面，具备以下功能：

在终端中运行Claude（或兼容模型），拥有完整的文件系统访问权限
通过Pipe IPC和局域网发现支持多实例协同
通过ACP协议与Zed、Cursor等IDE集成
提供企业级监控（Langfuse、Sentry、GrowthBook）
支持网页搜索、语音输入、电脑/浏览器控制
可与任何Anthropic兼容的API提供商（OpenAI、Gemini等）配合使用

Installation

安装方法

From NPM (Recommended)

从NPM安装（推荐）

bash

npm i -g claude-code-best

bash

npm i -g claude-code-best

Start CCB with Node.js

使用Node.js启动CCB

ccb

Start CCB with Bun runtime

使用Bun运行时启动CCB

ccb-bun

Update to latest version

更新至最新版本

ccb update

undefined

ccb update

undefined

From Source

从源码安装

Prerequisites:

bash

undefined

前置要求：

bash

undefined

Install Bun >= 1.3.11

安装Bun >= 1.3.11

curl -fsSL https://bun.sh/install | bash exec $SHELL # Reload shell

curl -fsSL https://bun.sh/install | bash exec $SHELL # 重新加载shell

Verify installation

验证安装

bun --version bun upgrade # Update if needed


**Clone and run:**
```bash
git clone https://github.com/claude-code-best/claude-code.git
cd claude-code
bun install

bun --version bun upgrade # 如有需要进行更新


**克隆并运行：**
```bash
git clone https://github.com/claude-code-best/claude-code.git
cd claude-code
bun install

Development mode (version 888)

开发模式（版本888）

bun run dev

Build for production

生产环境构建

bun run build

Run built version

运行构建后的版本

node dist/cli.js

or

或

bun dist/cli.js

undefined

bun dist/cli.js

undefined

Configuration

配置说明

Login to API Provider

登录API提供商

On first run, enter

/login

in the REPL:

bash

ccb

首次运行时，在REPL中输入

/login

：

bash

ccb

In REPL:

在REPL中：

/login


Choose provider type:
- **Anthropic Compatible** - any Claude-compatible API
- **OpenAI Compatible** - OpenAI or compatible services
- **Gemini Compatible** - Google Gemini API

**Required fields:**

| Field | Description | Example |
|-------|-------------|---------|
| Base URL | API endpoint | `https://api.anthropic.com` |
| API Key | Auth token | `$ANTHROPIC_API_KEY` |
| Haiku Model | Fast model | `claude-haiku-4-5-20251001` |
| Sonnet Model | Balanced model | `claude-sonnet-4-6` |
| Opus Model | Premium model | `claude-opus-4-6` |

Use **Tab/Shift+Tab** to navigate, **Enter** to confirm.

/login


选择提供商类型：
- **Anthropic兼容** - 任何Claude兼容的API
- **OpenAI兼容** - OpenAI或兼容服务
- **Gemini兼容** - Google Gemini API

**必填字段：**

| 字段 | 描述 | 示例 |
|-------|-------------|---------|
| Base URL | API端点 | `https://api.anthropic.com` |
| API Key | 认证令牌 | `$ANTHROPIC_API_KEY` |
| Haiku Model | 快速模型 | `claude-haiku-4-5-20251001` |
| Sonnet Model | 平衡模型 | `claude-sonnet-4-6` |
| Opus Model | 高级模型 | `claude-opus-4-6` |

使用**Tab/Shift+Tab**导航，**Enter**确认。

Environment Variables

环境变量

bash

undefined

bash

undefined

Enable features via FEATURE_<NAME>=1

通过FEATURE_<NAME>=1启用功能

FEATURE_BUDDY=1 ccb

Remote control

远程控制

CLAUDE_BRIDGE_BASE_URL=https://remote-control.claude-code-best.win/
CLAUDE_BRIDGE_OAUTH_TOKEN=$YOUR_TOKEN
ccb --remote-control

Langfuse monitoring

Langfuse监控

LANGFUSE_PUBLIC_KEY=$YOUR_KEY
LANGFUSE_SECRET_KEY=$YOUR_SECRET
LANGFUSE_HOST=https://cloud.langfuse.com
ccb

Poor mode (reduce API calls)

精简模式（减少API调用）

ccb # then type /poor in REPL

undefined

ccb # 然后在REPL中输入/poor

undefined

Key Commands

核心命令

In REPL

在REPL中

bash

/login              # Configure API provider
/pipes              # Show active pipe connections
/poor               # Toggle poor mode (reduce costs)
/dream              # Organize memory files
/voice doubao       # Enable voice input
/teach-me <topic>   # Interactive learning about CCB architecture

bash

/login              # 配置API提供商
/pipes              # 显示活跃的管道连接
/poor               # 切换精简模式（降低成本）
/dream              # 整理内存文件
/voice doubao       # 启用语音输入
/teach-me <topic>   # 交互式学习CCB架构

Examples:

示例：

/teach-me Claude Code architecture /teach-me React Ink terminal rendering --level beginner /teach-me Tool system --resume

undefined

/teach-me Claude Code architecture /teach-me React Ink terminal rendering --level beginner /teach-me Tool system --resume

undefined

CLI Flags

CLI参数

bash

undefined

bash

undefined

Start with remote control

启动远程控制

ccb --remote-control

Enable channels (external notifications)

启用频道（外部通知）

ccb --channels plugin:name@marketplace

Debug mode with inspector

带调试器的调试模式

bun run dev:inspect

Then attach VS Code debugger (F5 → "Attach to Bun (TUI debug)")

然后连接VS Code调试器（F5 → "Attach to Bun (TUI debug)"）

undefined

undefined

Usage Patterns

使用模式

Basic Usage

基础使用

typescript

// CCB is primarily a CLI tool, but internally uses these patterns:

// 1. Tool execution (from src/tools/)
import { executeShellCommand } from './tools/bash';

const result = await executeShellCommand({
  command: 'ls -la',
  workingDir: process.cwd()
});

// 2. Multi-instance coordination (Pipe IPC)
import { PipeManager } from './ipc/pipe-manager';

const pipeManager = new PipeManager();
await pipeManager.broadcast({
  type: 'message',
  content: 'Task complete'
});

// 3. Memory management
import { MemoryService } from './memory/service';

const memory = new MemoryService();
await memory.extract({
  sessionId: 'current',
  content: 'User prefers TypeScript over JavaScript'
});

typescript

// CCB主要是一款CLI工具，但内部使用以下模式：

// 1. 工具执行（来自src/tools/）
import { executeShellCommand } from './tools/bash';

const result = await executeShellCommand({
  command: 'ls -la',
  workingDir: process.cwd()
});

// 2. 多实例协同（Pipe IPC）
import { PipeManager } from './ipc/pipe-manager';

const pipeManager = new PipeManager();
await pipeManager.broadcast({
  type: 'message',
  content: 'Task complete'
});

// 3. 内存管理
import { MemoryService } from './memory/service';

const memory = new MemoryService();
await memory.extract({
  sessionId: 'current',
  content: 'User prefers TypeScript over JavaScript'
});

Remote Control Setup

远程控制设置

bash

undefined

bash

undefined

Self-hosted Docker setup

自托管Docker部署

docker run -d
-p 3000:3000
-e OAUTH_SECRET=$YOUR_SECRET
claude-code-remote-control

Client connection

客户端连接

CLAUDE_BRIDGE_BASE_URL=http://localhost:3000
CLAUDE_BRIDGE_OAUTH_TOKEN=$YOUR_TOKEN
ccb --remote-control

undefined

CLAUDE_BRIDGE_BASE_URL=http://localhost:3000
CLAUDE_BRIDGE_OAUTH_TOKEN=$YOUR_TOKEN
ccb --remote-control

undefined

Web Search Integration

网页搜索集成

typescript

// Enabled automatically when API keys are set
export BING_SEARCH_API_KEY=$YOUR_BING_KEY

typescript

undefined

or

设置API密钥后自动启用

export BRAVE_SEARCH_API_KEY=$YOUR_BRAVE_KEY

ccb

export BING_SEARCH_API_KEY=$YOUR_BING_KEY

In session, ask: "search for latest TypeScript features"

或

undefined

export BRAVE_SEARCH_API_KEY=$YOUR_BRAVE_KEY

ccb

Langfuse Monitoring

在会话中询问："search for latest TypeScript features"

bash

undefined

undefined

Configure via env vars

Langfuse监控

export LANGFUSE_PUBLIC_KEY=pk-lf-xxx export LANGFUSE_SECRET_KEY=sk-lf-xxx export LANGFUSE_HOST=https://cloud.langfuse.com

ccb

bash

undefined

Every agent loop will be tracked in Langfuse dashboard

通过环境变量配置

undefined

export LANGFUSE_PUBLIC_KEY=pk-lf-xxx export LANGFUSE_SECRET_KEY=sk-lf-xxx export LANGFUSE_HOST=https://cloud.langfuse.com

ccb

Computer Use (Screen Control)

每个代理循环都会在Langfuse仪表盘中被追踪

typescript

// Requires computer-use feature enabled
FEATURE_COMPUTER_USE=1 ccb

// In session:
// "take a screenshot of my screen"
// "click on the file menu"
// "type 'hello world' in the editor"

undefined

Chrome Automation

电脑控制（屏幕操控）

bash

undefined

typescript

undefined

Install chrome-use MCP server first

需要启用computer-use功能

npx @modelcontextprotocol/create-server chrome-use

FEATURE_COMPUTER_USE=1 ccb

Configure in CCB

在会话中：

—

"take a screenshot of my screen"

—

"click on the file menu"

—

"type 'hello world' in the editor"

ccb

undefined

In REPL: configure MCP server path

Chrome自动化

Usage in session:

—

"open chrome and navigate to github.com"

—

"fill out the login form with my credentials"

—

"scrape the table data from this page"

—

undefined

bash

undefined

Development Patterns

先安装chrome-use MCP服务器

Project Structure

—

claude-code/
├── src/
│   ├── cli.ts              # Entry point
│   ├── tools/              # Built-in tools (bash, read_file, etc.)
│   ├── ipc/                # Pipe IPC for multi-instance
│   ├── memory/             # Memory extraction & retrieval
│   ├── providers/          # API provider adapters
│   ├── ui/                 # React Ink terminal components
│   └── features/           # Feature flag implementations
├── dist/                   # Build output (450+ chunks)
├── build.ts                # Bun build script
└── package.json

npx @modelcontextprotocol/create-server chrome-use

Adding a Custom Tool

在CCB中配置

typescript

// src/tools/my-custom-tool.ts
import { z } from 'zod';

export const MyCustomToolSchema = z.object({
  input: z.string().describe('Tool input parameter')
});

export async function executeMyCustomTool(
  params: z.infer<typeof MyCustomToolSchema>
): Promise<string> {
  // Your tool logic
  return `Processed: ${params.input}`;
}

// Register in src/tools/index.ts
export const tools = [
  {
    name: 'my_custom_tool',
    description: 'Does something useful',
    input_schema: MyCustomToolSchema,
    execute: executeMyCustomTool
  },
  // ... other tools
];

ccb

Debugging in VS Code

在REPL中：配置MCP服务器路径

—

会话中的使用示例：

—

"open chrome and navigate to github.com"

—

"fill out the login form with my credentials"

—

"scrape the table data from this page"

json

// .vscode/launch.json (already configured)
{
  "configurations": [
    {
      "name": "Attach to Bun (TUI debug)",
      "type": "bun",
      "request": "attach",
      "url": "ws://localhost:8888"
    }
  ]
}

Steps:

Terminal:
```
bun run dev:inspect
```
Set breakpoints in
```
src/
```
VS Code: F5 → Select "Attach to Bun (TUI debug)"

undefined

Common Issues

开发模式

Installation Fails

项目结构

bash

undefined

claude-code/
├── src/
│   ├── cli.ts              # 入口文件
│   ├── tools/              # 内置工具（bash、read_file等）
│   ├── ipc/                # 多实例Pipe IPC
│   ├── memory/             # 内存提取与检索
│   ├── providers/          # API提供商适配器
│   ├── ui/                 # React Ink终端组件
│   └── features/           # 功能标志实现
├── dist/                   # 构建输出（450+个代码块）
├── build.ts                # Bun构建脚本
└── package.json

Clear old versions

添加自定义工具

npm rm -g claude-code-best

typescript

// src/tools/my-custom-tool.ts
import { z } from 'zod';

export const MyCustomToolSchema = z.object({
  input: z.string().describe('Tool input parameter')
});

export async function executeMyCustomTool(
  params: z.infer<typeof MyCustomToolSchema>
): Promise<string> {
  // 你的工具逻辑
  return `Processed: ${params.input}`;
}

// 在src/tools/index.ts中注册
export const tools = [
  {
    name: 'my_custom_tool',
    description: 'Does something useful',
    input_schema: MyCustomToolSchema,
    execute: executeMyCustomTool
  },
  // ... 其他工具
];

Install specific version

在VS Code中调试

npm i -g claude-code-best@5.0.0

json

// .vscode/launch.json（已配置）
{
  "configurations": [
    {
      "name": "Attach to Bun (TUI debug)",
      "type": "bun",
      "request": "attach",
      "url": "ws://localhost:8888"
    }
  ]
}

步骤：

终端：
```
bun run dev:inspect
```
在
```
src/
```
中设置断点
VS Code：F5 → 选择"Attach to Bun (TUI debug)"

If Bun install has issues, use npm

常见问题

—

安装失败

npm i -g claude-code-best

undefined

bash

undefined

Bun Not Found After Install

清除旧版本

bash

undefined

npm rm -g claude-code-best

macOS/Linux - reload shell

安装特定版本

exec $SHELL

npm i -g claude-code-best@5.0.0

Or manually source

如果Bun安装有问题，使用npm

source ~/.zshrc # or ~/.bashrc

npm i -g claude-code-best

undefined

Windows - reopen PowerShell

安装后找不到Bun

undefined

bash

undefined

Feature Not Working

macOS/Linux - 重新加载shell

bash

undefined

exec $SHELL

Check feature flags are set

或手动加载

FEATURE_NAME=1 ccb

source ~~/.zshrc # 或~~/.bashrc

Example for all features:

Windows - 重新打开PowerShell

FEATURE_BUDDY=1
FEATURE_FORK_SUBAGENT=1
FEATURE_WEB_SEARCH=1
ccb

undefined

undefined

API Connection Issues

功能无法使用

bash

undefined

bash

undefined

Verify credentials in REPL

检查功能标志是否已设置

/login

FEATURE_NAME=1 ccb

Re-enter Base URL and API Key

启用所有功能的示例：

Test with simple request

—

In session: "what's 2+2"

—

undefined

FEATURE_BUDDY=1
FEATURE_FORK_SUBAGENT=1
FEATURE_WEB_SEARCH=1
ccb

undefined

Build Errors

API连接问题

bash

undefined

bash

undefined

Update Bun to latest

在REPL中验证凭据

bun upgrade

/login

Clean install

重新输入Base URL和API Key

—

用简单请求测试

—

在会话中："what's 2+2"

rm -rf node_modules bun.lockb bun install

undefined

Rebuild

构建错误

bun run build

undefined

bash

undefined

Remote Control Not Connecting

更新Bun到最新版本

bash

undefined

bun upgrade

Verify env vars are exported

清理安装

echo $CLAUDE_BRIDGE_BASE_URL echo $CLAUDE_BRIDGE_OAUTH_TOKEN

rm -rf node_modules bun.lockb bun install

Test with official hosted version

重新构建

CLAUDE_BRIDGE_BASE_URL=https://remote-control.claude-code-best.win/
CLAUDE_BRIDGE_OAUTH_TOKEN=test-my-key
ccb --remote-control

undefined

bun run build

undefined

Advanced Configuration

远程控制无法连接

Multi-Instance Pipe IPC

—

bash

undefined

bash

undefined

Terminal 1 (main instance)

验证环境变量是否已导出

ccb

echo $CLAUDE_BRIDGE_BASE_URL echo $CLAUDE_BRIDGE_OAUTH_TOKEN

Terminal 2 (sub instance, auto-discovers main)

使用官方托管版本测试

ccb

CLAUDE_BRIDGE_BASE_URL=https://remote-control.claude-code-best.win/
CLAUDE_BRIDGE_OAUTH_TOKEN=test-my-key
ccb --remote-control

undefined

In main REPL:

高级配置

—

多实例Pipe IPC

/pipes # Shows connected instances

bash

undefined

Shift+↓ to broadcast message to all pipes

终端1（主实例）

undefined

ccb

LAN Cross-Machine Discovery

终端2（子实例，自动发现主实例）

bash

undefined

ccb

Machine A

在主REPL中：

FEATURE_LAN_DISCOVERY=1 ccb

/pipes # 显示已连接的实例

Machine B (same network)

Shift+↓ 向所有管道广播消息

FEATURE_LAN_DISCOVERY=1 ccb

undefined

Instances automatically discover and connect

局域网跨机器发现

undefined

bash

undefined

Channels (External Notifications)

机器A

bash

undefined

FEATURE_LAN_DISCOVERY=1 ccb

Install channel plugin (e.g., Slack)

机器B（同一网络）

ccb --channels plugin:slack@marketplace

FEATURE_LAN_DISCOVERY=1 ccb

Configure webhook in REPL

实例会自动发现并连接

Now external events can push messages into session

—

undefined

undefined

Custom Memory Optimization

频道（外部通知）

bash

undefined

bash

undefined

In REPL, manually trigger memory organization

安装频道插件（如Slack）

/dream

ccb --channels plugin:slack@marketplace

Or configure auto-dream

在REPL中配置webhook

—

现在外部事件可以将消息推送到会话中

FEATURE_AUTO_DREAM=1 ccb

undefined

undefined

Resources

自定义内存优化

Homepage: https://ccb.agent-aura.top/
Documentation: https://ccb.agent-aura.top/docs/
GitHub: https://github.com/claude-code-best/claude-code
Discord: https://discord.gg/uApuzJWGKX
Feature Docs: See
```
docs/features/
```
in repository

bash

undefined

Example Session

在REPL中手动触发内存整理

bash

$ ccb
Claude Code Best v5.0.0

You: I need to analyze all TypeScript files in src/ and create a dependency graph

/dream

—

或配置自动整理

—

FEATURE_AUTO_DREAM=1 ccb

undefined

—

资源

—

主页: https://ccb.agent-aura.top/
文档: https://ccb.agent-aura.top/docs/
GitHub: https://github.com/claude-code-best/claude-code
Discord: https://discord.gg/uApuzJWGKX
功能文档: 查看仓库中的
```
docs/features/
```
目录

—

示例会话

—

bash

$ ccb
Claude Code Best v5.0.0

You: I need to analyze all TypeScript files in src/ and create a dependency graph