JA

jorgejie/ai_scaffold

Developer tools
53 stars 품질 40 트렌드 40

ai_scaffold 是一套基于模板的脚手架系统,为你的现有项目生成完整的 AI 辅助开发框架。它通过**一套统一架构**支持 **4 种 AI 工具 × 5 种平台**。

개요

AI Coding Skeleton v2.0 — CLI + LLM 协作架构:CLI搭建骨架 → AI深度初始化 AI 编码骨架 v2.0 — 为你的项目生成结构化的规则、技能、Agent 和 Hook 体系。 ai_scaffold v2.0 is a that generates a complete AI coding assistance framework for your existing project. It uses a unique two-phase approach: 1. (10 seconds): Auto-detects your project type and creates the skeleton structure 2. (2-5 minutes): Scans your source code, understands architecture, and generates customized rules Instead of generating static templates with placeholders, ai_scaffold leverages AI to create rules that truly match your project's actual code structure.

README

ai_scaffold

博客地址点这里

AI Coding Skeleton v2.0 — CLI + LLM 协作架构:CLI搭建骨架 → AI深度初始化

AI 编码骨架 v2.0 — 为你的项目生成结构化的规则、技能、Agent 和 Hook 体系。

English | 中文


English

What is ai_scaffold?

ai_scaffold v2.0 is a CLI + LLM collaborative scaffolding system that generates a complete AI coding assistance framework for your existing project. It uses a unique two-phase approach:

  1. CLI Phase (10 seconds): Auto-detects your project type and creates the skeleton structure
  2. AI Phase (2-5 minutes): Scans your source code, understands architecture, and generates customized rules

Instead of generating static templates with placeholders, ai_scaffold leverages AI to create rules that truly match your project’s actual code structure.

Key Features

  • CLI + LLM Collaboration: CLI builds skeleton instantly, AI fills in customized content by scanning source code
  • Multi-tool support: Works with Claude Code, Qoder, Codex, and OpenCode — same content, different “shell”
  • Multi-platform: Android, iOS, HarmonyOS, Flutter, React Native (and generic fallback)
  • SKILL.md as AI Manual: Comprehensive initialization guide that AI follows to generate project-specific rules
  • Cross-platform rule templates: Platform-specific rules auto-injected based on actual project analysis
  • Split-file architecture: Rules, skills, agents, hooks, and references are loaded on-demand — no bloated single file
  • Conflict resolution: Explicit priority-based arbitration when multiple rules collide
  • Auto-generated references: AI scans project structure + reads source code to generate accurate module documentation
  • Proactive error correction: An agent that actively scans code compliance, not just passively reviews
  • CodeGraph integration (optional): Dynamically detects CodeGraph CLI — when present, references/ operates in lightweight mode while CodeGraph handles structural exploration, reducing AI tool calls by ~90%

Architecture

                    npx ai-scaffold-pro (CLI Phase - 10s)
                               │
          ┌────────────────────┼────────────────────┐
          ▼                    ▼                    ▼
   ┌──────────────┐   ┌──────────────┐   ┌──────────────────┐
   │  detect.js   │   │promptMinimal │   │ createSkeleton   │
   │              │   │   Config()   │   │                  │
   │  Platform    │   │              │   │  Create dirs     │
   │  Build Sys   │──▶│ Language +   │──▶│  Copy base files │
   │  Language    │   │ AI Tool      │   │  Generate minimal│
   │  NDK/C++     │   │ selection    │   │  entry file      │
   │  CodeGraph   │   │ (2 questions)│   │  + AI_INIT_GUIDE │
   └──────────────┘   └──────────────┘   └────────┬─────────┘
                                                   │
                                    Transaction-safe: temp → validate → atomic commit
                                                   │
                                                   ▼
                              Skeleton Created! Now switch to AI Phase ↓

              User opens project in AI tool (Claude Code/Qoder/etc.)
                               │
                    User: "Initialize using SKILL.md"
                               │
                        AI Phase (2-5 minutes)
                               │
          ┌────────────────────┼────────────────────┐
          ▼                    ▼                    ▼
   ┌──────────────┐   ┌──────────────┐   ┌──────────────────┐
   │  Phase 0-1   │   │  Phase 2     │   │   Phase 3        │
   │              │   │              │   │                  │
   │ Scan project │──▶│ Interactive  │──▶│ Generate custom  │
   │ structure    │   │ config       │   │ content:         │
   │ Read source  │   │ (7-11 Qs)    │   │ - rules/         │
   │ Analyze arch │   │              │   │ - agents/        │
   │              │   │              │   │ - skills/        │
   │              │   │              │   │ - references/    │
   └──────────────┘   └──────────────┘   └────────┬─────────┘
                                                   │
                                            Validation & Done ✅

Quick Start

Step 1: CLI Skeleton Creation (10 seconds)

# Run directly with npx (no install needed)
cd your-project/
npx ai-scaffold-pro

# Follow the minimal prompts:
# 1. Select language (中文 / English)
# 2. Select AI tool (Claude Code / Qoder / Codex / OpenCode)
# 3. Project auto-detected
# 4. Skeleton created!

Step 2: AI Deep Initialization (2-5 minutes)

# Open your project in your chosen AI tool:
# - Claude Code
# - Qoder
# - Codex
# - OpenCode

# Then enter this command in the AI chat:
"Initialize using .qoder/skills/project_initialization/SKILL.md"

# Or simply:
"按 project_initialization skill 初始化本项目"

The AI will automatically:

  1. Scan your project structure
  2. Read key source files
  3. Ask 7-11 configuration questions
  4. Generate customized rules based on actual code
  5. Create comprehensive references documentation
  6. Set up intelligent agents and hooks

Total time: ~3 minutes (10s CLI + 2-5min AI)

Supported Platforms

Platform Build System Detection File Rule Templates
Android Gradle settings.gradle / build.gradle.kts MVVM, ARouter, Retrofit, ViewBinding, Paging 3
iOS Xcode / SPM *.xcodeproj / Package.swift SwiftUI/UIKit, ARC memory, Swift concurrency, App Store compliance
HarmonyOS Hvigor hvigor-config.json5 Stage model, ArkUI, ArkTS state management, router
Flutter Dart Pub pubspec.yaml Widget tree, Provider/Riverpod/BLoC, GoRouter, platform channels
React Native npm/yarn package.json + react-native TurboModules/JSI bridge, StyleSheet, React Navigation, Hermes
Generic — — Fallback for any project

Comparison

ai_scaffold Superkit Superpowers create-*-app
Focus AI Coding rule skeleton AI full-stack starter AI capability plugins Project scaffolding
Value Multi-tool x multi-platform rule generation Pre-built AI-integrated templates Extend AI tool abilities Standardized project init
Best for Adding AI assistance to existing projects Starting new projects from scratch Enhancing AI tool capabilities New project standardization

Project Structure (After AI Initialization)

{{ENTRY}}                          ← Entry point (full version, generated by AI)
AI_INIT_GUIDE.md                   ← User-friendly initialization guide
{{DIR}}/
├── CHANGELOG.md                   ← Version tracking
├── settings.json                  ← Hook registration
├── settings.local.json            ← Local permissions
├── hooks/
│   ├── post-edit-tracker.sh       ← Edit tracking
│   └── check-review-needed.sh     ← Review reminder
├── rules/
│   ├── project_rule.md            ← Main project rules (AI-generated)
│   └── conflict_resolution.md     ← Conflict arbitration (AI-generated)
├── skills/
│   ├── plan_mode/SKILL.md         ← Task planning (AI-generated)
│   ├── code_review/SKILL.md       ← Code review (AI-generated)
│   ├── performance_check/SKILL.md ← Performance & security audit (AI-generated)
│   └── project_initialization/    ← **AI initialization manual**
│       └── SKILL.md               ← This is what AI follows!
├── agents/
│   ├── arch-review.md             ← Architecture review (AI-generated)
│   ├── resource-sync.md           ← Resource sync check (AI-generated)
│   ├── proactive-correction.md    ← Proactive error correction (AI-generated)
│   └── cpp-memory-review.md       ← C++ memory safety (NDK only, AI-generated)
├── scripts/
│   └── gen_references.py          ← Structure scanner (--lightweight with CodeGraph)
└── references/
    ├── _scan.json                 ← Scan intermediate data
    ├── dependencies.md            ← Dependency graph (AI-generated from source)
    ├── conventions.md             ← Coding conventions (AI-generated from source)
    └── {module}.md × N            ← Module docs (AI reads source code to generate)

Note: Files marked “AI-generated” are created during Phase 2 (AI initialization), not by the CLI.


中文

ai_scaffold 是什么?

ai_scaffold v2.0 是一套 CLI + LLM 协作的脚手架系统,为你的现有项目生成完整的 AI 辅助开发框架。它采用独特的两阶段方法:

  1. CLI 阶段(10秒):自动检测项目类型并创建骨架结构
  2. AI 阶段(2-5分钟):扫描源代码、理解架构、生成定制化规则

不同于生成带有占位符的静态模板,ai_scaffold 利用 AI 创建真正匹配你项目实际代码结构的规则。

核心特性

  • CLI + LLM 协作:CLI 瞬间搭建骨架,AI 通过扫描源码填充定制化内容
  • 多工具支持:Claude Code、Qoder、Codex、OpenCode — 内容完全一致,仅"外壳"不同
  • 多平台适配:Android、iOS、鸿蒙、Flutter、React Native(及通用降级)
  • SKILL.md 作为 AI 手册:全面的初始化指南,AI 遵循它生成项目专属规则
  • 跨平台规则模板库:根据实际项目分析自动注入平台专属规则
  • 分包架构:规则、技能、Agent、Hook、参考资料按需加载,无臃肿单文件
  • 冲突裁决:多规则碰撞时有显式的优先级仲裁机制
  • 自动生成 References:AI 扫描项目结构 + 精读源码,生成准确的模块文档
  • 主动纠错:Agent 主动扫描代码合规性,而非被动审查
  • 性能与安全检查:自动检测内存泄漏、OOM风险、ANR、卡顿、安全漏洞(由 proactive-correction 调用)
  • CodeGraph 集成(可选):自动检测 CodeGraph CLI——安装后 references/ 采用轻量模式,结构探索由 CodeGraph 负责,相比手动扫描代码可减少约 90% 的 AI 工具调用

快速开始

步骤 1:CLI 骨架搭建(10秒)

# 直接用 npx 运行(无需安装)
cd your-project/
npx ai-scaffold-pro

# 按最小化提示操作:
# 1. 选择语言(中文 / English)
# 2. 选择 AI 工具(Claude Code / Qoder / Codex / OpenCode)
# 3. 自动检测项目
# 4. 骨架创建完成!

步骤 2:AI 深度初始化(2-5分钟)

# 在你选择的 AI 工具中打开项目:
# - Claude Code
# - Qoder
# - Codex
# - OpenCode

# 然后在 AI 对话框中输入:
"按 .qoder/skills/project_initialization/SKILL.md 初始化"

# 或者简单说:
"按 project_initialization skill 初始化本项目"

AI 将自动执行:

  1. 扫描项目结构
  2. 读取关键源文件
  3. 询问 7-11 个配置问题
  4. 基于实际代码生成定制化规则
  5. 创建完整的 references 文档
  6. 设置智能 Agent 和 Hook

总耗时:约3分钟(10秒 CLI + 2-5分钟 AI)

项目结构(AI 初始化后)

{{ENTRY}}                          ← 入口调度层(完整版,由 AI 生成)
AI_INIT_GUIDE.md                   ← 用户友好的初始化指南
{{DIR}}/
├── CHANGELOG.md                   ← 版本记录
├── settings.json                  ← Hook 注册
├── settings.local.json            ← 本地权限
├── hooks/
│   ├── post-edit-tracker.sh       ← 编辑追踪
│   └── check-review-needed.sh     ← 审查提醒
├── rules/
│   ├── project_rule.md            ← 项目主规则(AI 生成)
│   └── conflict_resolution.md     ← 冲突裁决(AI 生成)
├── skills/
│   ├── plan_mode/SKILL.md         ← 任务规划(AI 生成)
│   ├── code_review/SKILL.md       ← 代码审查(AI 生成)
│   ├── performance_check/SKILL.md ← 性能与安全检查(AI 生成)
│   └── project_initialization/    ← **AI 初始化手册**
│       └── SKILL.md               ← AI 就是按照这个执行的!
├── agents/
│   ├── arch-review.md             ← 架构审查(AI 生成)
│   ├── resource-sync.md           ← 资源同步(AI 生成)
│   ├── proactive-correction.md    ← 主动纠错(AI 生成)
│   └── cpp-memory-review.md       ← C++ 内存安全(仅 NDK,AI 生成)
├── scripts/
│   └── gen_references.py          ← 结构扫描器(--lightweight 配合 CodeGraph)
└── references/
    ├── _scan.json                 ← 扫描中间数据
    ├── dependencies.md            ← 依赖关系图(AI 从源码生成)
    ├── conventions.md             ← 编码约定(AI 从源码生成)
    └── {module}.md × N            ← 模块文档(AI 读取源码生成)

注意:标记为"AI 生成"的文件是在阶段 2(AI 初始化)中创建的,不是由 CLI 生成的。

支持的平台

平台 构建系统 检测文件 规则模板
Android Gradle settings.gradle / build.gradle.kts MVVM, ARouter路由, Retrofit网络, ViewBinding, Paging3
iOS Xcode / SPM *.xcodeproj / Package.swift SwiftUI/UIKit, ARC内存, Swift并发, App Store合规
鸿蒙 Hvigor hvigor-config.json5 Stage模型, ArkUI组件, ArkTS状态管理, router路由
Flutter Dart Pub pubspec.yaml Widget树, Provider/Riverpod/BLoC, GoRouter, Platform Channel
React Native npm/yarn package.json + react-native TurboModules/JSI桥接, StyleSheet, React Navigation, Hermes
通用 — — 任意项目的兜底方案

Contributing

See CONTRIBUTING.md for guidelines.

License

MIT

More Information

  • 📖 REFACTORING.md - Detailed explanation of v2.0 architecture changes
  • 📝 docs/article2.md - In-depth article about ai_scaffold design philosophy
View this README on GitHub

추천 도구

다른 키워드를 입력하거나 필터를 제거해 보세요.

설치

npx skillfish add jorgejie/ai_scaffold