Three-Tomato

将需求文档转换为多端原生开发代码，支持 Android、iOS、鸿蒙、小程序、快应用、H5、Web、macOS、Windows 等 17 个平台。

核心原则：原生优先，AI 友好技术栈（数据集大、文档丰富）

Overview

This skill enables AI Agents to analyze requirement documents and generate native platform-specific code using AI-friendly tech stacks with large training datasets.

Supported Platforms

Trigger Commands

Code Generation

🤖 "transform to [platform]"
🤖 "generate [platform] code"
🤖 "convert requirement to multi-platform"
🤖 "create platform specs"
🤖 "转换为[平台]代码"
🤖 "生成多端代码"

Self-Evolution

🧬 "3T evolve" / "3T 总结归纳进化"
🧬 "3T log issue [description]" / "3T 记录问题 [描述]"
🧬 "3T show evolution log" / "3T 查看进化日志"

Workflow

Evolution Workflow (3T evolve / 3T 总结归纳进化)

When 3T evolve command is triggered, AI Agent executes:

Step 1: Collect Feedback

1. Scan .three-tomato/feedback/issues/ for issue records
2. Analyze compile/runtime errors in .three-tomato/output/
3. Identify patterns from user’s manual code fixes

Step 2: Categorize Issues

1. Group by platform (Android/iOS/Mini Programs, etc.)
2. Group by type (syntax error/API changes/best practices/architecture)
3. Calculate issue frequency

Step 3: Update Knowledge

1. Update references/patterns/known-issues.yaml – known issues database
2. Update references/patterns/best-practices.yaml – best practices
3. Update platform-specific plugins/[platform]-generator/PLUGIN.md
4. Log changes to .three-tomato/evolution/changelog.md

Step 4: Validate Improvements

1. Generate test cases for high-frequency issues
2. Verify improved templates resolve the issues
3. Output evolution report to .three-tomato/evolution/reports/

Phase 1: Requirement Analysis

1. Read requirement document from .three-tomato/requirements/ or specified path
2. Parse and extract:
   • Feature specifications
   • UI/UX requirements
   • Data models
   • API interfaces
   • Business logic flows

Phase 2: Platform Selection

1. Read .three-tomato/config.yaml for target platforms
2. Check enabled plugins in plugins/_registry.yaml
3. Determine output platforms based on user command or config

Phase 3: Code Generation

For each target platform:

1. Load platform-specific templates from references/templates/[platform]/
2. Apply tech stack configurations
3. Generate:
   • Project structure
   • UI components
   • Data models
   • API clients
   • Business logic
   • Platform-specific configurations

Phase 4: Output & Validation

1. Write generated code to .three-tomato/output/[platform]/
2. Generate comparison matrix
3. Create migration guides between platforms

Configuration

Read configuration from .three-tomato/config.yaml:

# Target platforms to generate
platforms:
  - android
  - ios
  - wechat-mp
  - h5

# AI-Friendly Tech Stack (原生优先，数据集大)
tech_stack:
  android:
    language: kotlin              # AI 数据集大
    ui: compose                   # 声明式，AI 友好
    architecture: mvvm
    di: hilt
    network: retrofit + okhttp
    async: coroutines
    
  ios:
    language: swift               # 现代语法，AI 熟悉
    ui: swiftui                   # 声明式 UI
    architecture: mvvm
    network: urlsession           # 原生优先
    async: async-await
    
  harmony:
    language: arkts               # TypeScript 超集
    ui: arkui
    architecture: mvvm
    
  wechat-mp:
    language: typescript          # 类型安全，AI 友好
    framework: native             # 原生开发
    
  alipay-mp:
    language: typescript
    framework: native
    
  baidu-mp:
    language: typescript
    framework: native
    
  quick-app:
    language: typescript
    
  h5:
    language: typescript
    framework: react              # React 优先，数据集最大
    ui: antd-mobile
    bundler: vite
    
  web:
    language: typescript
    framework: react              # React 优先
    meta_framework: nextjs        # SSR/SSG 支持
    ui: antd / shadcn-ui
    state: zustand
    
  macos:
    language: swift
    ui: swiftui
    architecture: mvvm
    
  windows:
    language: csharp              # C# AI 数据集大
    ui: winui3                    # 现代 Windows UI
    architecture: mvvm

# Output settings
output:
  directory: .three-tomato/output
  include_tests: true
  include_docs: true
  language: zh-CN

Plugin System

Hook Points

Plugins can extend functionality at these hooks:

• before_analyze – Pre-process requirement documents
• after_analyze – Post-process extracted requirements
• before_generate – Modify generation parameters
• on_generate – Custom code generation logic
• after_generate – Post-process generated code
• on_export – Custom export formats

Plugin Commands

📋 "list platform plugins"
📦 "install plugin <source>"
✅ "enable plugin <name>"
❌ "disable plugin <name>"

Built-in Plugins

Output Structure

.three-tomato/
├── config.yaml              # Main configuration
├── requirements/            # Input requirement documents
│   ├── PRD.md              # Product requirement document
│   ├── api.yaml            # API specifications
│   └── ui-specs/           # UI design specs
├── cache/                   # Incremental cache
├── output/                  # Generated code
│   ├── android/            # Kotlin + Compose
│   ├── ios/                # Swift + SwiftUI
│   ├── harmony/            # ArkTS + ArkUI
│   ├── wechat-mp/          # TypeScript 原生
│   ├── alipay-mp/          # TypeScript 原生
│   ├── baidu-mp/           # TypeScript 原生
│   ├── quick-app/          # TypeScript
│   ├── h5/                 # React + TypeScript
│   ├── web/                # Next.js + React
│   ├── macos/              # Swift + SwiftUI
│   ├── windows/            # C# + WinUI 3
│   └── _shared/            # 共享资源
│       ├── models/         # 数据模型定义
│       ├── api/            # API 接口定义
│       └── assets/         # 共享资源文件
├── reports/                 # Analysis reports
│   ├── comparison.md       # Platform comparison
│   ├── migration.md        # Migration guide
│   └── compatibility.md    # Compatibility matrix
└── i18n/                    # Multi-language docs
    ├── en/
    └── zh/

Important Instructions

For AI Agent:

1. Always read config first: Check .three-tomato/config.yaml before generation
2. Respect tech stack: Use configured frameworks and libraries
3. Maintain consistency: Ensure data models and API calls are consistent across platforms
4. Platform idioms: Follow each platform’s best practices and conventions
5. Protocol Retention:
   • If requirement/legacy uses gRPC:
     • Supported Platforms (iOS, Android, Windows, macOS, Flutter): MUST use gRPC to maintain high performance and type safety.
     • Restricted Platforms (Mini Programs, H5): Fallback to HTTP/JSON (via Envoy/Gateway).
   • If requirement uses REST/GraphQL: Use standard HTTP clients.
6. Incremental updates: Only regenerate changed parts using cache
7. Preserve user content: Content marked with <!-- user-content --> must not be overwritten

Code Quality Standards:

• Follow platform-specific coding guidelines
• Include proper error handling
• Add necessary comments (in configured language)
• Generate unit tests when include_tests: true
• Create platform-specific README with setup instructions

Cross-Platform Consistency:

• Unified data model definitions
• Consistent API interface naming
• Shared business logic documentation
• Synchronized i18n resources

References

• Templates: references/templates/
• Prompts: references/prompts/
• Examples: references/examples/
• Platform Guides: docs/platforms/

Version

• Skill Version: 1.0.1
• Last Updated: 2026-01-30
• Author: three-tomato