MCPxelMCPxel
Co-Pilot / 辅助式
更新于 a month ago

ai-dev-standards

Ddaffy0208
0.0k
daffy0208/ai-dev-standards
82
Agent 评分

💡 摘要

一个全面的框架，用于AI辅助的软件开发，具有自动验证和代理评估功能。

🎯 适合人群

AI开发者软件工程师产品经理数据科学家DevOps专业人士

🤖 AI 吐槽: “看起来很能打，但别让配置把人劝退。”

安全分析中风险

风险：Medium。建议检查：是否执行 shell/命令行指令；是否发起外网请求（SSRF/数据外发）；API Key/Token 的获取、存储与泄露风险；文件读写范围与路径穿越风险；依赖锁定与供应链风险。以最小权限运行，并在生产环境启用前审计代码与依赖。

AI Development Standards

CI Status License: MIT TypeScript Node

Version 3.1.0 | Last Updated: 2025-11-24

A comprehensive framework of specialized AI skills, MCP servers, and development tools for AI-assisted software development. Features automated validation, agent evaluation, and quality assurance.

📊 Current Resources

Total: 199 Resources

| Category | Count | Description | | ---------------- | ----- | ------------------------------------------------- | | Skills | 64 | Specialized AI methodologies and workflows | | MCPs | 51 | Model Context Protocol servers (executable tools) | | Tools | 4 | Core utility scripts | | Components | 75 | Reusable UI and system components | | Integrations | 5 | Third-party service connectors |

MCP Coverage: 79.7% (51 MCPs supporting 64 Skills)

✨ What's New in 3.1.0

🎯 Agent Evaluation System (Phase 5.12)

Implement Eval-Driven Development (EDD) for continuous agent quality assurance:

  • Automated testing against golden datasets
  • Multiple grading strategies (exact, regex, LLM-based)
  • Performance metrics and regression tracking
  • 100% test pass rate in validation suite
# Run agent evaluations
node scripts/run-agent-evals.js --dataset tests/fixtures/golden-dataset-example.json --mock

⚡ Two-Tier Validation System

  • Quick Validation (10-30s): Registry consistency, documentation
  • Full Validation (2-5min): Includes linting, type checking, tests, agent evals
npm run validate:quick  # Fast feedback
npm run validate:full   # Comprehensive checks

📚 Enhanced Documentation

  • .claude/CLAUDE.md - Complete Claude Code configuration guide
  • FINAL-RESOURCE-COUNTS.md - Resource tracking and metrics
  • docs/VALIDATION-SYSTEM.md - Validation methodology

🚀 Quick Start

📖 New to this repository? Check out our Installation Guide and Quick Start Guide for step-by-step instructions.

For New Projects

# Clone the repository
git clone https://github.com/daffy0208/ai-dev-standards.git
cd ai-dev-standards

# Install dependencies
npm install

# Run validation to ensure everything works
npm run validate

For Existing Projects

# Clone as a reference
git clone https://github.com/daffy0208/ai-dev-standards.git ~/ai-dev-standards

# Reference skills and patterns in your .cursorrules or .claude/claude.md
# See docs/EXISTING-PROJECTS.md for integration guide

Using with Claude Code

  1. Open your project in Claude Code

  2. Reference this repository in your project instructions:

    You have access to ai-dev-standards at ~/ai-dev-standards

When needed, reference skills from skills/ and patterns from standards/
Use the skill-registry.json to find relevant skills for tasks

  3. Claude will automatically discover and use appropriate skills

📖 What This Repository Does

Think of this as a shared knowledge base between you and Claude:

🎓 64 Specialized Skills

Methodologies Claude follows automatically:

  • Product: mvp-builder, product-strategist, go-to-market-planner
  • AI/ML: rag-implementer, multi-agent-architect, knowledge-graph-builder
  • Development: frontend-builder, api-designer, backend-architect
  • Infrastructure: deployment-advisor, security-engineer, performance-optimizer
  • Design: ux-designer, visual-designer, design-system-architect
  • Quality: testing-strategist, quality-auditor, agent-evaluator

🔧 51 MCP Servers

Executable tools that extend Claude's capabilities:

  • Search: semantic-search-mcp, dark-matter-analyzer-mcp
  • Quality: code-quality-scanner-mcp, security-scanner-mcp, test-runner-mcp
  • AI/Data: vector-database-mcp, embedding-generator-mcp, knowledge-base-mcp
  • Design: figma-sync-mcp, design-token-manager-mcp, theme-builder-mcp
  • DevOps: deployment-orchestrator-mcp, database-migration-mcp

📐 Architecture Patterns

Proven approaches for complex systems:

  • RAG architectures (Naive, Advanced, Modular)
  • Multi-agent coordination patterns
  • Event-driven systems
  • Real-time data pipelines
  • Authentication patterns

🛡️ Quality Assurance

  • Automated validation system (2-tier)
  • Agent evaluation framework (EDD)
  • Security best practices
  • Performance standards
  • Accessibility guidelines

💡 Key Features

⚡ Automated Validation

# Quick validation (10-30 seconds)
npm run validate:quick

# Full validation (2-5 minutes)
npm run validate:full

# Agent evaluation only
node scripts/run-agent-evals.js --dataset tests/fixtures/golden-dataset-example.json --mock

Validates:

  • ✅ Registry consistency
  • ✅ Documentation accuracy
  • ✅ Code quality (ESLint)
  • ✅ Type safety (TypeScript)
  • ✅ Test coverage
  • ✅ Agent performance (NEW)

🤖 Agent Evaluation System

Test AI agents against golden datasets to ensure consistent, high-quality outputs:

{
  "tests": [
    {
      "id": "T001",
      "input": "Create a React button component with TypeScript",
      "expected": "import React from 'react';",
      "grading": { "type": "contains", "threshold": 0.8 }
    }
  ]
}

Features:

  • Multiple grading types (exact match, contains, regex, LLM-graded)
  • Performance metrics (latency, success rate, score)
  • Historical tracking and regression detection
  • Custom dataset support

📊 Comprehensive Documentation

  • For Developers: docs/GETTING-STARTED.md, docs/QUICK-START.md
  • For AI: meta/PROJECT-CONTEXT.md, meta/HOW-TO-USE.md
  • Configuration: .claude/CLAUDE.md, FINAL-RESOURCE-COUNTS.md
  • Validation: docs/VALIDATION-SYSTEM.md

🎯 Smart Resource Discovery

# Find skills for a task
grep -r "mvp" meta/skill-registry.json

# Search all resources
grep -r "authentication" meta/

# View resource counts
cat FINAL-RESOURCE-COUNTS.md

🗂️ Repository Structure

ai-dev-standards/
├── skills/                    # 64 specialized methodologies
│   ├── mvp-builder/          # MVP development & prioritization
│   ├── rag-implementer/      # RAG system implementation
│   ├── api-designer/         # API design patterns
│   └── [61 more...]
│
├── mcp-servers/              # 51 executable tools
│   ├── semantic-search-mcp/  # Semantic code search
│   ├── vector-database-mcp/  # Vector DB integration
│   ├── code-quality-scanner-mcp/
│   └── [48 more...]
│
├── standards/                # Architecture & best practices
│   ├── architecture-patterns/
│   ├── best-practices/
│   ├── coding-conventions/
│   └── project-structure/
│
├── meta/                     # Resource registry & context
│   ├── registry.json         # Master resource registry
│   ├── skill-registry.json   # Skill catalog
│   ├── mcp-registry.json     # MCP catalog
│   └── PROJECT-CONTEXT.md    # For AI assistants
│
├── docs/                     # Comprehensive documentation
│   ├── GETTING-STARTED.md
│   ├── VALIDATION-SYSTEM.md
│   ├── AGENT-VALIDATION.md   # NEW!
│   └── [40+ more guides...]
│
├── scripts/                  # Automation & validation
│   ├── run-agent-evals.js    # NEW! Agent evaluation
│   ├── validate-full.sh      # Full validation suite
│   └── [20+ more scripts...]
│
└── tests/                    # Test suites & fixtures
    ├── fixtures/
    │   └── golden-dataset-example.json  # NEW!
    └── [150+ test files...]

🎯 Usage Examples

Example 1: Starting a New Project

User: "I want to build a SaaS product for invoice management"

Claude uses:
1. product-strategist → Validate problem-solution fit
2. mvp-builder → Identify P0 features (invoicing, payment tracking)
3. frontend-builder → React/Next.js structure
4. api-designer → REST API design
5. deployment-advisor → Vercel + Railway recommendation
6. security-engineer → Auth, data encryption, PCI compliance

Example 2: Implementing AI Search

User: "Add AI-powered search to our documentation"

Claude uses:
1. rag-implementer → RAG methodology
2. rag-pattern.md → Advanced RAG architecture
3. vector-database-mcp → Pinecone integration
4. embedding-generator-mcp → OpenAI embeddings
5. semantic-search-mcp → Search implementation

Example 3: Code Quality Audit

User: "Audit our codebase for quality issues"

Claude uses:
1. quality-auditor → Comprehensive audit methodology
2. code-quality-scanner-mcp → Static analysis
3. security-scanner-mcp → Vulnerability detection
4. performance-profiler-mcp → Performance bottlenecks
5. test-runner-mcp → Test coverage analysis
6. agent-evaluator → AI agent quality checks (NEW!)

🔍 Finding Skills

By Task

# Search skills by keyword
grep -i "authentication" meta/skill-registry.json
grep -i "database" meta/skill-registry.json
grep -i "testing" meta/skill-registry.json

By Category

View meta/skill-registry.json for complete categorization:

  • Product & Business (8 skills)
  • AI & Machine Learning (10 skills)
  • Frontend Development (6 skills)
  • Backend Development (8 skills)
  • Infrastructure & DevOps (8 skills)
  • Design & UX (12 skills)
  • Quality & Testing (12 skills)

Auto-Discovery

Skills activate automatically based on your conversation with Claude. Just describe what you want to build!

⚙️ Validation System

Two-Tier Approach

Tier 1: Quick Validation (10-30 seconds)

npm run validate:quick

Checks:

  • Registry consistency
  • Documentation accurac
五维分析
清晰度8/10
创新性7/10
实用性9/10
完整性9/10
可维护性8/10
优缺点分析

优点

  • 全面的验证系统
  • 专业的AI技能
  • 自动化的代理评估
  • 丰富的文档

缺点

  • 初学者的复杂设置
  • 文档可能令人不知所措
  • 需要对AI概念的熟悉
  • 依赖外部工具

相关技能

useful-ai-prompts

A
toolCo-Pilot / 辅助式
88/ 100

“一个提示的宝藏，但别指望它们为你写小说。”

查看测评

mcpspy

A
toolCo-Pilot / 辅助式
86/ 100

“MCPSpy：因为谁不想窥探他们 AI 的秘密？”

查看测评

fastmcp

A
toolCo-Pilot / 辅助式
86/ 100

“FastMCP：因为谁不喜欢在AI中添加一点复杂性呢？”

查看测评

免责声明：本内容来源于 GitHub 开源项目，仅供展示和评分分析使用。

版权归原作者所有 daffy0208.