CC-Best

Role-Driven Development Workflow for Claude Code

English | 中文 | 🌐 Homepage

Transform Claude into a complete development team.
From product requirements to code review — one plugin, full workflow.

44 commands · 19 skills · 8 agents · 43 rules · 30 hooks

Quick Start • Features • Workflow • Commands • FAQ

Why CC-Best?

🎯 The Problem: Claude Code is powerful, but configuring workflows, coding standards, and safety rules from scratch takes hours.

✨ The Solution: Pre-configured roles (PM → Lead → Designer → Dev → QA) that mirror real team collaboration, with safety guardrails built-in.

⚡ Quick Start

# Add marketplace and install
/plugin marketplace add xiaobei930/cc-best
/plugin install cc-best@xiaobei930

# Verify installation (30 seconds)
/cc-best:status

# Start using
/cc-best:iterate "implement user authentication"

💡 Enable Hooks: After installation, run /cc-best:setup --hooks to activate safety guards and automation hooks.
See Hooks Configuration for details.

Plugin vs Clone: Command Format

💡 Recommended: Install via plugin for automatic updates and easier management.
All documentation uses plugin format (/cc-best:xxx). Clone users: run convert-to-local.js.

git clone https://github.com/xiaobei930/cc-best.git .claude
cd .claude && bash scripts/shell/init.sh

# Convert command format from /cc-best:xxx to /xxx
node scripts/node/convert-to-local.js

Use clone when you need to customize all files in your repo.

⚠️ Important: All documentation uses plugin format (/cc-best:xxx).
Run the conversion script to update to local format (/xxx).

# Remove the plugin
/plugin uninstall cc-best@xiaobei930

# Remove marketplace (optional)
/plugin marketplace remove xiaobei930/cc-best

No files are left behind.

What Happens Next?

After running /cc-best:iterate "implement user authentication", Claude will:

1. 📋 /cc-best:pm     → Analyze requirements, create task breakdown
2. 🏗️ /cc-best:lead   → Design technical solution
3. 💻 /cc-best:dev    → Write code, create tests
4. 🧪 /cc-best:qa     → Run tests, verify quality
5. ✅ /cc-best:commit → Commit changes with proper message

You just watch. Intervene only when needed.

Common Workflows

Who Is This For?

✨ Core Features

While CC-Best is built for Claude Code, the methodology (Dao-Fa-Shu-Qi) and role-driven patterns are framework-agnostic and can be adapted for other AI coding assistants.

What Makes CC-Best Different

Not just a prompt template collection — CC-Best simulates real team collaboration:

• 7 roles with clear boundaries: PM → Lead → Designer → Dev → QA → Verify → Commit
• Each role has explicit MUST/SHOULD/NEVER rules, output templates, and handoff protocols
• Automatic flow: PM creates REQ → Lead reviews & creates DES/TSK → Dev implements → QA validates
• Downstream correction (A3): Lead can adjust PM decisions; QA distinguishes implementation bugs from requirement assumption errors
• Document traceability: REQ-XXX → DES-XXX → TSK-XXX numbered chain

/cc-best:iterate enables fully autonomous development:

Read progress.md → Select role → Execute → Verify → Commit → Next task (no waiting)

• Smart role selection: 8 state conditions determine which role activates
• A1-A5 decision principles: Context inference (A1), decision recording (A2), downstream correction (A3), MVP fallback (A4), issue classification (A5)
• 4 strict stop conditions: All tasks done, user interrupt, fatal error, external dependency
• Cross-session continuity: memory-bank + progress.md rolling window

CC-Best learns from your development patterns:

observe → analyze → learn → evolve

• observe: observe-patterns.js hook automatically tracks tool usage patterns
• analyze: /cc-best:analyze mines git history and usage data
• learn: /cc-best:learn extracts reusable knowledge
• evolve: /cc-best:evolve generates new commands, skills, or agents from learned patterns

📁 Directory Structure

your-project/
├── CLAUDE.md          # Project constitution
├── commands/          # 44 slash commands
├── skills/            # 19 development skills
├── agents/            # 8 specialized agents
├── rules/             # 43 coding standards (10 dirs)
├── hooks/             # Safety hooks
├── scripts/           # Automation (node/python/shell)
├── memory-bank/       # Progress & architecture docs
└── .claude/           # Claude Code config

🔄 Workflow

Standard Development Cycle

flowchart LR
    PM["/pm<br/>Requirements"] --> Clarify["/clarify<br/>Clarify"]
    Clarify --> Lead["/lead<br/>Design"]
    Lead --> Designer["/designer<br/>UI Design"]
    Designer --> Dev["/dev<br/>Implement"]
    Dev --> QA["/qa<br/>Test"]
    QA --> Verify["/verify<br/>Verify"]
    Verify --> Commit["/commit<br/>Commit"]
    Commit --> Clear["/clear<br/>Clear Context"]
    Clear -.->|Loop| PM

Three Development Modes

Core behavior: Task complete → Update progress.md → Read next task → Execute immediately (no waiting).

Stop conditions: All tasks done | User interrupt (Ctrl+C) | Fatal error | External dependency needed.

5 mandatory confirmation checkpoints:

Learning mode: /cc-best:pair --learn "teach me unit testing" — Claude explains every step in detail.

Safe autonomy: Even in pair mode, Claude can freely read files, search code, run tests, and format code.

📖 Detailed usage guide: See MODES.md for comprehensive documentation on each mode, including when to use, how to control, and best practices.

📋 Command Reference

44 commands organized into categories:

📖 Full reference: See COMMANDS.md for all parameters and usage examples.

🛠️ Skills

19 development skills organized by domain:

📖 Full reference: See skills/README for detailed skill documentation.

🏗️ Architecture Overview

CC-Best uses a four-tier architecture:

flowchart TB
    subgraph User["👤 You"]
        CMD["/cc-best:iterate 'add feature'"]
    end

    subgraph Commands["📋 Commands (44)"]
        PM["/pm"] --> Lead["/lead"] --> Dev["/dev"] --> QA["/qa"]
    end

    subgraph Skills["🛠️ Skills (19)"]
        S1["backend · frontend · testing · security"]
        S2["architecture · devops · git"]
        S3["learning · compact · exploration"]
    end

    subgraph Agents["🤖 Agents (8)"]
        A1["architect · planner · code-reviewer"]
        A2["code-simplifier · security-reviewer"]
        A3["tdd-guide · build-error-resolver · requirement-validator"]
    end

    subgraph Safety["🛡️ Safety Hooks (18)"]
        H1["PreToolUse: validate, secrets, protect"]
        H2["PostToolUse: format, typecheck, observe"]
    end

    CMD --> Commands
    Commands -.->|"auto-inject"| Skills
    Commands -.->|"delegate"| Agents
    Commands -.->|"guard"| Safety

8 specialized agents: architect, build-error-resolver, code-reviewer, code-simplifier, planner, requirement-validator, security-reviewer, tdd-guide

📐 Full documentation: See ARCHITECTURE.md for component relationships and call chains.

🤖 Agent details: See agents/README for agent capabilities and invocation.

🔌 Plugin Compatibility

CC-Best is designed to work seamlessly with official Claude Code plugins. Our built-in agents and skills complement (not replace) official plugins.

Relationship with Official Plugins

Recommended Plugin Configuration

{
  "enabledPlugins": {
    "code-review@claude-plugins-official": true,
    "hookify@claude-plugins-official": true,
    "security-guidance@claude-plugins-official": true
  }
}

When to Use Which

• No plugins installed: Built-in agents/skills work standalone
• With plugins: Use official plugins for advanced features, CC-Best for quick local checks
• Best practice: Install official plugins, use built-in agents for immediate feedback, official plugins for deep analysis

⚙️ Customization Guide

Adding New Rules

Create a file in rules/:

---
paths:
  - "**/*.your-ext"
---

# Rule Title

## Rule Content

...

Adding New Commands

Create a file in commands/:

---
allowed_tools:
  - Read
  - Edit
  - Write
  - Bash
---

# /your-command - Command Name

## Responsibilities

...

## Execution Steps

1. ...
2. ...

Configuring Hooks

Edit .claude/settings.local.json:

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "Bash",
        "hooks": [
          {
            "type": "command",
            "command": "python scripts/your-script.py",
            "timeout": 5
          }
        ]
      }
    ]
  }
}

Pre-configured Hooks

⚠️ Important: Hooks Require Manual Setup

Due to a known Claude Code issue, plugin hooks are disabled by default. To enable the safety guards and automation hooks, run:

/cc-best:setup --hooks

This command configures hooks with absolute paths. See the FAQ for more details.

All hooks default to Node.js for cross-platform compatibility. Python/Bash alternatives available in scripts/.

💡 Best Practices

1. Keep CLAUDE.md Concise

• Keep it under 100 lines
• Put detailed specifications in rules/

2. Use the Memory Bank

• Update progress.md after each task completion
• Record important decisions in architecture.md

3. Context Management

• Normal mode: Use /clear frequently to avoid context overflow
• /cc-best:iterate mode: Don't clear manually, maintain loop continuity

4. Don't Overload MCP

• Enable no more than 10 MCP servers per project
• Use disabledMcpServers to disable unused ones

5. Regular Cleanup

• Delete unused language rules
• Remove unused commands

6. MCP Temporary Directory Management

MCP tools auto-create temporary directories in your project:

Cleanup Script: Use cleanup.sh for regular maintenance:

# Preview files to delete (dry run)
bash scripts/shell/cleanup.sh --dry-run

# Clean files older than 7 days (default)
bash scripts/shell/cleanup.sh

# Clean files older than 3 days
bash scripts/shell/cleanup.sh --days 3

# Clean all MCP temporary files
bash scripts/shell/cleanup.sh --all

❓ FAQ

📖 Full FAQ: See FAQ.md for comprehensive troubleshooting guides.

Quick Answers

Run /cc-best:setup --verify to diagnose. Common fixes:

• Clone users: cp .claude/settings.local.json.example .claude/settings.local.json
• Plugin users: Run /cc-best:setup --hooks to configure absolute paths
• Windows: See FAQ.md for ${CLAUDE_PLUGIN_ROOT} workarounds

• Interrupt: Press Ctrl+C (or Esc in some terminals)
• Pause: Type anything — Claude will wait for your input
• Resume: Just continue the conversation

Claude saves progress to memory-bank/progress.md, so you can always resume later.

Claude will:

1. Analyze the failure
2. Return to /cc-best:dev to fix the issue
3. Re-run /cc-best:qa to verify

If stuck after 3 attempts, Claude will ask for your input. You can:

• Provide hints: "Try checking the database connection"
• Skip the test: "Skip this test for now"
• Take over: "I'll fix this manually"

Edit .claude/settings.local.json:

{ "enabledMcpjsonServers": ["memory", "sequential-thinking"] }

Best practice: Enable ≤10 MCP servers per project.

Both are excellent. Choose based on your needs:

💡 They can coexist! Use CC-Best for workflows, Superpowers for git automation.

🔧 Requirements

Optional MCP Servers

Some commands use MCP (Model Context Protocol) tools for enhanced functionality:

Note: These are optional. Commands work without MCP servers but with reduced functionality.
Install via Claude Code settings: Settings > MCP Servers

Supported Languages

📚 References

Official Resources

• Anthropic Official Best Practices
• CLAUDE.md Complete Guide

Plugin Documentation

• Quick Start Guide - Get started in 5 minutes
• Advanced Guide - Deep dive into methodology and architecture

🌐 Note on Internal Files

CC-Best's internal files (commands/, rules/, skills/) are written in Chinese. This is intentional:

• Claude understands Chinese - All Claude models can read and follow Chinese instructions perfectly
• No translation burden - Maintaining dual-language internal files would be impractical
• Focus on users - The README (this file) is fully English for international users

If you prefer English internal files, community contributions for English translations are welcome!

🤝 Contributing

Contributions are welcome! See CONTRIBUTING.md for details.

📄 License

MIT License - Free to use and modify

🌐 Visit Homepage · ⭐ Star on GitHub

If CC-Best helps you, please give it a ⭐ Star!

Star History