new-project（新项目）

Initialize New Project

IMPORTANT: Every project gets the same skeleton structure, but the CONTENT (architecture, MCP servers, skills, requirements) is unique and must be researched.

Phase 1: Create Skeleton Structure (ALWAYS THE SAME)

mkdir -p {{PROJECT_PATH}}/{docs,.claude/skills}

Create these files immediately (with placeholders):

project/
├── CLAUDE.md                    # Project context
├── .mcp.json                    # MCP servers (empty initially)
├── .claude/
│   ├── settings.json            # Hooks and permissions
│   └── skills/                  # Custom commands
└── docs/
    ├── PRD.md                   # Product requirements
    ├── ARCHITECTURE.md          # System design
    └── REQUIREMENTS.md          # Functional specs

Phase 2: Gather Project Context

Ask user:

1. Project name and one-line description
2. Problem being solved - What pain point?
3. Target users - Who will use this?
4. Key features - What must it do?
5. Constraints - Budget, timeline, existing systems?

Phase 3: Research MCP Servers (CRITICAL)

Based on project needs, research available MCP servers:

Search for MCP servers:

• Web search: "MCP server for [technology/service]"
• Check: https://github.com/modelcontextprotocol/servers
• Check: https://mcp.so/ (MCP registry)

Common MCP servers by domain:

For each potential server, evaluate:

• Does it solve a real need for this project?
• What credentials/setup does it require?
• Will it be used frequently enough to justify the context cost?

Phase 4: Research Skills Needed

Identify repetitive workflows that should become skills:

Universal skills (most projects need):

• build - Compile/bundle the application
• test - Run test suite
• deploy - Push to production

Research project-specific skills:

• What multi-step processes will be repeated?
• What commands are complex and error-prone?
• What workflows need standardization?

Example skill patterns by project type:

Phase 5: Research Architecture Options

Don't assume architecture - research what fits:

Questions to answer:

1. What scale is expected? (users, requests, data volume)
2. What's the budget constraint?
3. What does the team already know?
4. What are similar products built with?

Web search for:

• "[problem domain] architecture patterns"
• "[similar product] tech stack"
• "best practices [technology] 2025"

Phase 6: Fill Planning Documents

With research complete, populate:

CLAUDE.md

• Project overview and goals
• Chosen tech stack with rationale
• MCP servers configured and why
• Available skills and when to use them
• Common commands
• Code standards for this project

docs/PRD.md

• User personas from Phase 2
• Feature requirements with priorities
• Success metrics
• Timeline and milestones

docs/ARCHITECTURE.md

• System diagram
• Data models
• API design
• Infrastructure plan

docs/REQUIREMENTS.md

• Detailed functional requirements (REQ-XXX-NNN format)
• Non-functional requirements (performance, security)

Phase 7: Configure MCP Servers

Create .mcp.json with researched servers:

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "./"]
    }
    // Add other researched servers
  }
}

Phase 8: Create Initial Skills

For each identified skill, create .claude/skills/{name}/SKILL.md:

---
name: skill-name
description: When to use this skill. Be specific.
---

# Skill Name

Steps to execute...

Phase 9: Initialize Git

cd {{PROJECT_PATH}}
git init
cat > .gitignore << 'EOF'
node_modules/
.env*
.next/
out/
dist/
*.tfstate*
.terraform/
.claude/session.log
.claude/settings.local.json
EOF
git add -A
git commit -m "Initial project setup with Claude Code structure"

Output

Display:

1. Project structure tree
2. Configured MCP servers with purpose
3. Created skills with descriptions
4. Next steps:

- Add MCP server credentials to environment - Run /mcp to verify server connections - Review and refine PRD with stakeholders - Begin implementation per architecture plan