Claude Skills Repository

A curated collection of reusable skills for Claude Code. Pick the skills you need and add them to your project to enhance Claude's capabilities.

🎯 What is This?

This repository provides ready-to-use skills that extend Claude Code's functionality. Each skill is a self-contained module that teaches Claude how to perform specific tasks or follow particular workflows in your projects.

📦 Available Skills

Skill Name	Description	Source	Installation
create-skill-file	Guides Claude in creating well-structured SKILL.md files with templates, examples, and best practices	-	cp -r create-skill-file .claude/skills/
prompt-optimize	Optimize your prompt with Claude	-	cp -r prompt-optimize .claude/skills/
deep-reading-analyst	Comprehensive framework for deep analysis using 10+ thinking models (SCQA, 5W2H, Critical Thinking, Mental Models, First Principles, etc.)	🔗 GitHub	cp -r deep-reading-analyst .claude/skills/
dry-refactoring	Systematic code refactoring following DRY principle with 4-step workflow to eliminate code duplication	-	cp -r dry-refactoring .claude/skills/
frontend-design	Creates unique, production-grade frontend interfaces with exceptional design quality and creative aesthetics	-	cp -r frontend-design .claude/skills/
mcp-builder	Guide for creating high-quality MCP servers that enable LLMs to interact with external services through tools	-	cp -r mcp-builder .claude/skills/
daily-ai-news	Aggregates and summarizes the latest AI news from multiple sources with concise briefs and direct links	-	cp -r daily-ai-news .claude/skills/
fastgpt-workflow-generator	Generates production-ready FastGPT workflow JSON from natural language requirements with AI-powered template matching and three-layer validation	-	cp -r fastgpt-workflow-generator .claude/skills/
planning-with-files	Manus-style workflow using persistent markdown files for planning, progress tracking, and knowledge storage with 3-file pattern	🔗 GitHub	cp -r planning-with-files .claude/skills/
local-diff-review	Pre-PR local code review skill based on git diff, applies the same three-dimension standards as pr-review (code quality, style guide, common issues checklist)	-	cp -r local-diff-review .claude/skills/

Total: 10 skills available

---

Skill Details

🔧 create-skill-file

Versions: Chinese / English

A meta-skill that teaches you how to create high-quality SKILL.md files for Claude.

What's included:

• ✅ Comprehensive writing guidelines
• ✅ Ready-to-use templates (Basic & Workflow)
• ✅ Real-world examples (Good & Bad practices)
• ✅ Quality checklist and troubleshooting guide

Trigger Keywords: "create skill", "write skill", "SKILL.md", "skill guidelines", "best practices"

Installation:

# Chinese version
cp -r create-skill-file .claude/skills/

# English version
cp -r create-skill-file-EN .claude/skills/

---

🔧 prompt-optimize

Version: Chinese

An expert prompt engineering skill that transforms Claude into "Alpha-Prompt" - a master prompt engineer who collaboratively crafts high-quality prompts through flexible dialogue.

What's included:

• ✅ Expert prompt engineering consultation
• ✅ Advanced cognitive architectures (CoT, ToT, Self-Consistency, ReAct)
• ✅ Security guardrails and safety considerations
• ✅ Architecture upgrade suggestions for simple requirements
• ✅ Collaborative dialogue-based prompt optimization

Key Features:

• Flexible Communication: Genuine two-way dialogue, not rigid templated questions
• Proactive Architecture Upgrades: Suggests advanced techniques like Tree of Thought for creative tasks
• Security Awareness: Provides safety recommendations for public-facing AI roles
• Quality Standards: Delivers production-ready prompts with clear role definitions and structured outputs

Trigger Keywords: "optimize prompt", "improve prompt", "enhance AI instruction", "prompt engineering", "system instruction"

Installation:

cp -r prompt-optimize .claude/skills/

---

🔧 deep-reading-analyst

A professional skill that transforms surface-level reading into deep learning through systematic analysis using 10+ proven thinking frameworks.

For detailed documentation, see: Deep Reading Analyst Introduction

🔗 GitHub Repository: https://github.com/ginobefun/deep-reading-analyst-skill/tree/main

Installation:

cp -r deep-reading-analyst .claude/skills/

---

🔧 dry-refactoring

Version: Chinese

A systematic skill that guides code refactoring following the DRY (Don't Repeat Yourself) principle through a proven 4-step workflow.

What's included:

• ✅ Step 1: Identify Repetition (obvious and semantic duplication)
• ✅ Step 2: Abstract the Logic (functions, classes, constants)
• ✅ Step 3: Replace Implementation (systematic replacement)
• ✅ Step 4: Verify and Test (unit tests, integration tests, performance)

Key Features:

• Comprehensive Coverage: Handles copy-paste code, magic numbers, structural and logical repetition
• Step-by-Step Process: Clear workflow from identification to verification
• Real-World Examples: Complete e-commerce discount calculation refactoring case
• Best Practices: Includes common pitfalls, testing strategies, and gradual refactoring approach

Trigger Keywords: "DRY", "code duplication", "refactor repetitive code", "eliminate duplication", "magic numbers", "code smell", "extract function"

Installation:

cp -r dry-refactoring .claude/skills/

---

🔧 frontend-design

Version: Chinese

An expert skill for creating unique, production-grade frontend interfaces with exceptional design quality that avoids generic AI aesthetics.

What's included:

• ✅ Design thinking framework (Purpose, Style, Constraints, Differentiation)
• ✅ Typography guide (unique font selection, avoid common fonts)
• ✅ Color & theme systems (CSS variables, accessibility)
• ✅ Animation best practices (CSS animations, Framer Motion)
• ✅ Spatial composition techniques (asymmetric layouts, overlapping elements)
• ✅ Background & visual details (gradients, noise textures, glass morphism)

Key Features:

• Bold Aesthetic Choices: Guides selection of extreme, intentional design styles
• Anti-Generic AI Design: Explicitly avoids overused fonts (Inter, Roboto) and cliché color schemes
• Implementation Complexity Matching: Complex designs get complex code, minimal designs get precise code
• Production-Ready Code: Semantic HTML, accessibility, responsive design, performance optimization

Trigger Keywords: "web component", "landing page", "dashboard", "React component", "UI design", "beautify", "frontend"

Installation:

cp -r frontend-design .claude/skills/

---

🔧 mcp-builder

Version: English

A comprehensive guide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools.

What's included:

• ✅ Phase 1: Deep Research and Planning (MCP design patterns, API coverage)
• ✅ Phase 2: Tool Design (naming conventions, parameter schemas, error handling)
• ✅ Phase 3: Implementation (FastMCP/Python, MCP SDK/TypeScript)
• ✅ Phase 4: Testing and Publishing (integration tests, documentation)

Key Features:

• Modern MCP Design Patterns: Balances comprehensive API coverage with specialized workflow tools
• Framework-Specific Guidance: Detailed examples for both FastMCP (Python) and MCP SDK (TypeScript)
• Quality Standards: Emphasizes clear naming, actionable error messages, and proper context management
• Complete Workflow: From initial research through testing and publishing

Supported Frameworks:

• FastMCP (Python) - Recommended for rapid development
• MCP SDK (TypeScript/Node.js) - For JavaScript ecosystem integration

Trigger Keywords: "MCP server", "Model Context Protocol", "build MCP", "FastMCP", "MCP SDK", "tool integration"

Installation:

cp -r mcp-builder .claude/skills/

---

🔧 daily-ai-news

Version: Chinese & English

An intelligent skill that aggregates and summarizes the latest AI news from multiple sources (AI news websites and web search) with concise briefs and direct links to original articles.

What's included:

• ✅ Phase 1: Information Gathering (WebSearch + mcp__web_reader__webReader)
• ✅ Phase 2: Content Filtering (last 24-48 hours, deduplication)
• ✅ Phase 3: Categorization (Major Announcements, Research, Industry, Tools, Policy)
• ✅ Phase 4: Output Formatting (structured template with links)
• ✅ Comprehensive news sources database (20+ sources)
• ✅ Search query templates by category
• ✅ Multiple output format templates

Key Features:

• Multi-Source Aggregation: Fetches from 3-5 major AI news sites plus web search
• Smart Filtering: Keeps only recent news (24-48 hours) with major significance
• Structured Categorization: 5 categories - Major Announcements, Research & Papers, Industry & Business, Tools & Applications, Policy & Ethics
• Direct Links: Every news item includes a direct link to the original article
• Customization Options: Focus areas, depth levels, time ranges, and format preferences
• Bilingual Support: Works with both Chinese and English news sources

Trigger Keywords: "给我今天的AI资讯", "today's AI news", "AI updates", "latest AI developments", "daily AI briefing", "AI industry news", "artificial intelligence news"

Installation:

cp -r daily-ai-news .claude/skills/

---

🔧 fastgpt-workflow-generator

Version: English

An intelligent skill that automatically generates production-ready FastGPT workflow JSON from natural language requirements using AI-powered semantic template matching and comprehensive three-layer validation.

What's included:

• ✅ Phase 1: Requirements Analysis (AI semantic extraction)
• ✅ Phase 2: Template Matching (two-stage: coarse + fine filtering)
• ✅ Phase 3: JSON Generation (NodeId generation, auto-layout, reference handling)
• ✅ Phase 4: Validation (format, connections, logic completeness)
• ✅ Phase 5: Incremental Modification (add/remove/modify nodes)
• ✅ Built-in templates (document translation, sales training, resume screening, financial news)
• ✅ Complete references (40+ node types, validation rules, JSON structure specs)
• ✅ Validation script (Node.js)

Key Features:

• AI-Powered Template Matching: Two-stage semantic matching (metadata + AI analysis) to find the most similar template
• Automatic JSON Generation: Generates complete workflow with semantic NodeIds, auto-layout positions, and proper references
• Three-Layer Validation: Format (JSON structure) → Connections (node references) → Logic (workflow completeness)
• Incremental Modification Support: Add, delete, or modify nodes in existing workflows
• Built-in Templates: 4 production-ready templates covering different domains and complexities
• Portable Design: Self-contained with relative paths, works in any project

Built-in Templates:

• 文档翻译助手.json - Simple workflow (document processing)
• 销售陪练大师.json - Medium complexity (conversational AI)
• 简历筛选助手_飞书.json - Complex workflow (data + external integration)
• AI金融日报.json - Scheduled trigger + multi-agent (news aggregation)

Reference Format:

• Array reference: ["nodeId", "key"] - for direct values
• Template reference: {{$nodeId.key$}} - for string interpolation (double braces!)

Trigger Keywords: "create FastGPT workflow", "generate workflow JSON", "design FastGPT application", "工作流", "workflow automation", "multi-agent systems", "FastGPT templates"

Installation:

cp -r fastgpt-workflow-generator .claude/skills/

Quick Start:

# Example 1: Create from natural language
"生成一个旅游规划的工作流，收集用户的目的地、天数、预算等信息，然后AI生成详细的旅游规划"

# Example 2: Modify existing workflow
"在现有问答工作流前添加知识库搜索节点"

# Example 3: Validate workflow
"验证这个workflow JSON是否正确"

---

🔧 planning-with-files

Version: English

A Manus-style workflow skill that transforms how you work with Claude by using persistent markdown files as "working memory on disk" for planning, progress tracking, and knowledge storage.

🔗 Reference: Based on OthmanAdi/planning-with-files

What's included:

• ✅ 3-File Pattern System (task_plan.md, notes.md, deliverable.md)
• ✅ Task plan template with phases, decisions, and errors tracking
• ✅ Notes template for research and findings
• ✅ Progress tracking with checkboxes and status updates
• ✅ Knowledge persistence outside attention window
• ✅ Complete workflow loop patterns
• ✅ Real-world examples and anti-patterns

Key Features:

• Persistent Memory: Store plans, research, and decisions in files instead of context window
• 3-File Pattern: Systematic approach with task_plan.md (progress), notes.md (research), and deliverable.md (output)
• Structured Tracking: Phase-based planning with checkboxes, decisions log, and error tracking
• Loop-Based Workflow: Read → Work → Update cycle keeps goals fresh in attention
• Context Management: Prevents context stuffing by storing information in files
• Multi-Session Support: Persist work across conversation sessions

The 3-File Pattern:

• task_plan.md - Track phases and progress (update after each phase)
• notes.md - Store findings and research (during research)
• [deliverable].md - Final output (at completion)

Core Workflow Loop:

1. Create task_plan.md with goal and phases
2. Research → save to notes.md → update task_plan.md
3. Read notes.md → create deliverable → update task_plan.md
4. Deliver final output

When to Use:

• Starting complex multi-step projects
• Research tasks requiring information gathering
• Tasks requiring progress tracking across sessions
• User mentions "planning", "organizing work", "tracking progress"
• Need structured output with clear phases

Trigger Keywords: "complex task", "multi-step project", "planning", "organize work", "track progress", "research task", "structured output"

Installation:

cp -r planning-with-files .claude/skills/

Quick Start:

# Example: Plan a feature implementation
"I need to implement a new authentication system"

→ Claude creates:
  - task_plan.md (phases: design, implementation, testing, deployment)
  - notes.md (stores API research and design decisions)
  - Updates plan after each phase completion

---

🔧 local-diff-review

Version: Chinese

A pre-PR self-review skill that runs git diff locally and applies the same three-dimension review standards as a formal PR review — no custom weakening of criteria allowed.

What's included:

• ✅ Code quality standards (code-quality-standards.md)
• ✅ Common issues checklist (common-issues-checklist.md)
• ✅ 5 diff scope modes (unstaged, staged, HEAD, branch-vs-baseline, specific commit)
• ✅ Three-level issue grading (🔴 Critical / 🟡 Suggested / 🟢 Optional)
• ✅ Structured output: change overview → issue list → overall conclusion

Key Features:

• Review First, Fix Later: Outputs an issue list by default; only modifies code when explicitly asked
• Actionable Results: Every issue includes file path, line number, risk explanation, and a concrete fix suggestion
• Regression Risk Summary: Highlights i18n, type safety, edge cases, compatibility, and performance impact
• Consistent Standards: Strictly reuses the same criteria as pr-review; no softened rules

Trigger Keywords: "review 我本地改动", "local diff review", "review 我当前改动", "git diff review", "暂存区 review", "分支差异审查", "提 PR 前 review"

Installation:

cp -r local-diff-review .claude/skills/

Quick Start:

# Example 1: Review all unstaged changes
"请按 PR 标准 review 我本地当前改动。"

# Example 2: Review only staged changes
"请 review 我的 git diff --staged，只给问题清单。"

# Example 3: Review branch diff against baseline
"请按 FastGPT 规范审查我当前分支相对 upstream/v4.14.7-dev 的改动。"

---

🚀 How to Use

Installation Steps

1. Navigate to your project

   cd /path/to/your/project
   

2. Create the skills directory (if not exists)

   mkdir -p .claude/skills
   

3. Copy the skill you want

   Example - Adding the skill creation guide:

   # clone this repository if you haven't already
   git clone https://github.com/YYH211/Claude-meta-skill.git
   
   # For Chinese version
   cp -r /Claude-meta-skill/create-skill-file .claude/skills/
   
   # OR for English version
   cp -r /Claude-meta-skill/create-skill-file-EN .claude/skills/
   

4. Verify installation

   ls .claude/skills/
   

   You should see your copied skill directory:

   .claude/skills/
   ├── create-skill-file/    (or create-skill-file-EN/)
   │   ├── SKILL.md
   │   ├── templates/
   │   └── examples/
   

Verifying Configuration

✅ Method 1: Check File Structure

cat .claude/skills/create-skill-file/SKILL.md | head -10

Expected output should show the YAML frontmatter:

---
name: create-skill-file
description: Guides Claude in creating well-structured SKILL.md files...
---

✅ Method 2: Test with Claude

Start Claude Code and try triggering the skill:

claude

Then ask:

• "Help me create a new skill"
• "I want to write a SKILL.md file"
• "Guide me on skill best practices"

If Claude provides structured guidance referencing templates and examples, the skill is working! 🎉

✅ Method 3: List All Skills

You can check what skills are available by asking Claude:

What skills do you have access to?

claude response：

  1. create-skill-file (Skill 文件创建指南)

  功能: 指导创建结构良好的 SKILL.md 文件

  触发场景:
  - 当你询问如何创建 Skill
  - 需要编写 SKILL.md 文档
  - 想了解 Skill 编写最佳实践

  核心能力:
  - 提供清晰的命名、结构和内容组织指南
  - 包含模板和示例
  - 质量检查清单
  - 常见问题解答

  位置: .claude/skill/create-skill-file/SKILL.md

Or manually check:

find .claude/skills -name "SKILL.md" -exec grep -A 2 "^name:" {} \;

---

💡 Usage Examples

Example 1: Creating a New Skill

You: "I need to create a skill for managing Docker deployments"

Claude: [Uses create-skill-file skill]
"I'll help you create a Docker deployment skill. Let's start by following
the skill creation guidelines..."

Example 2: Improving Existing Skills

You: "Can you review my database-migration skill and suggest improvements?"

Claude: [References best practices from create-skill-file]
"Let me review your skill against the best practices checklist..."

---

🛠️ Troubleshooting

Skill Not Working

Problem	Solution
Claude doesn't use the skill	Verify the skill is in .claude/skills/ directory
Skill file not found	Check that SKILL.md exists and has valid frontmatter
Wrong skill activates	Make description more specific with unique keywords
Skill seems ignored	Try using explicit trigger keywords from the description

Common Issues

Q: Do I need to restart Claude after adding a skill?
A: Typically yes. Restart your Claude Code session after adding new skills.

Q: Can I modify the skills?
A: Absolutely! Feel free to customize them for your project needs.

Q: How many skills can I have?
A: No hard limit, but keep it manageable (5-10 skills per project is typical).

Q: Can I use multiple skills together?
A: Yes! Claude can use multiple skills in the same conversation as needed.

---

🚧 Coming Soon

This repository will grow with more useful skills:

• [ ] code-review-workflow - Systematic code review process
• [ ] api-documentation-generator - Generate API docs from code
• [ ] testing-strategy - Guide for writing comprehensive tests
• [ ] deployment-checklist - Pre-deployment verification
• [ ] refactoring-guide - Code refactoring best practices

Want to contribute? Create a skill following the guidelines in create-skill-file and submit it!

---

📚 Resources

• Claude Code Documentation
• Agent Skills Official Guide
• Best Practices

---

⭐ Star History

---

🤝 Partner Projects

FastGPT

FastGPT is a knowledge-based platform built on the LLM, offers out-of-the-box data processing and model invocation capabilities, allows for workflow orchestration through Flow visualization.

---

🤝 Contributing

Have a useful skill to share? We'd love to include it!

1. Create your skill following the create-skill-file guidelines
2. Test it thoroughly in real projects
3. Include clear documentation and examples
4. Submit via pull request or issue

---

📄 License

Free to use for any purpose. Customize and adapt as needed for your projects.

---

Happy coding with Claude! 🚀