CC-Best

---

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

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

Installation	Command format	Example
Plugin ⭐	/cc-best:command	/cc-best:iterate "add feature"
Clone	/command	/iterate "add feature"

💡 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

I want to...	Command	What Claude does
Build a feature	/cc-best:iterate "add dark mode toggle"	Full cycle: plan → code → test → commit
Fix a bug	/cc-best:iterate "fix login timeout issue"	Investigate → fix → verify → commit
Review code	/cc-best:pm "review recent changes"	Analyze code, suggest improvements
Learn the codebase	/cc-best:pair	Step-by-step exploration with explanations

Who Is This For?

You are...	Recommended mode	Why
Team with multiple roles	Full workflow	PM → Lead → Dev → QA mirrors your process
Solo developer	/cc-best:iterate	Let Claude handle the boring parts
Learning Claude Code	/cc-best:pair	Understand each step before proceeding
Need quick fixes	Direct commands	/cc-best:dev "fix the typo" for simple tasks

---

✨ Core Features

Feature	What it does
🎭 Role-Based Workflow	PM → Lead → Designer → Dev → QA — complete development cycle
🔄 Autonomous Mode	/cc-best:iterate runs tasks without intervention until completion
🛡️ Safety Hooks	Blocks rm -rf /, git push --force, and other risky commands
📐 Multi-Language Standards	8-dir layered structure: common + Python/frontend/Java/C#/C++/embedded/UI rules
🧠 Memory Bank	Persists progress and decisions across sessions
👥 Pair Programming	/cc-best:pair — step-by-step collaboration with 5 confirmation checkpoints
🔗 Knowledge Pipeline	observe → analyze → learn → evolve — self-improving knowledge loop
🌐 Cross-Platform	Windows, macOS, Linux — auto-detects package manager

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

Directory	Contents
commands/	Role commands (pm, lead, dev, qa), Mode commands (iterate, pair), Tool commands (build, test, commit)
skills/	Backend, Frontend, Testing, Security, DevOps, Architecture, Git
agents/	architect, build-error-resolver, code-reviewer, code-simplifier, planner, requirement-validator, security-reviewer, tdd-guide
rules/	43 rules in 10 dirs: common/ + python/, frontend/, java/, csharp/, cpp/, embedded/, ui/, rust/, go/
scripts/	Cross-platform hooks in Node.js (default), with Python/Bash alternatives
memory-bank/	progress.md (rolling window), architecture.md, tech-stack.md

---

🔄 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

Mode	Command	Use Case	Characteristics
Autonomous Iteration	/cc-best:iterate	Clear task list	Fully autonomous, no intervention needed
Pair Programming	/cc-best:pair	Learning, sensitive operations	Confirm each step, human-machine collaboration
Long-Running Loop	/cc-best:cc-ralph	Hour-level batch tasks	Requires ralph-loop plugin (/plugin install ralph-loop@claude-plugins-official)

Current State	Role Selected	Action
No requirements doc	/cc-best:pm	Requirement analysis
REQ has low-confidence items	/cc-best:clarify	Requirement clarification
Has REQ, no design	/cc-best:lead	Technical design
Has design, frontend tasks	/cc-best:designer	UI design guidance
Has tasks to implement	/cc-best:dev	Coding implementation
Code ready for verification	/cc-best:verify	Build + type + lint + test + security
Verification passed	/cc-best:qa	Functional acceptance

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:

Checkpoint	Example
Understanding	"I understand you need X. Correct?"
Design choice	"Option A or B? I recommend A because..."
Destructive action	"About to delete X. Confirm?"
External call	"Will call production API. Proceed?"
Commit	"Commit message: '...'. OK?"

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:

Category	Commands	Purpose
Role	/cc-best:pm, /cc-best:lead, /cc-best:dev, /cc-best:qa, /cc-best:designer, /cc-best:clarify, /cc-best:verify	Development workflow roles
Mode	/cc-best:iterate, /cc-best:pair, /cc-best:cc-ralph, /cc-best:mode, /cc-best:model	Autonomous/pair modes & model strategy
Build	/cc-best:build, /cc-best:test, /cc-best:run, /cc-best:fix	Build and test automation
Git	/cc-best:commit, /cc-best:pr, /cc-best:git-guide	Version control
Context	/cc-best:compact-context, /cc-best:checkpoint, /cc-best:catchup, /cc-best:context, /cc-best:memory	Session management
Quality	/cc-best:cleanup, /cc-best:docs, /cc-best:learn, /cc-best:analyze, /cc-best:evolve	Code quality & knowledge
Ops	/cc-best:fix-issue, /cc-best:release, /cc-best:service, /cc-best:hotfix	Issue fix, release, hotfix, services
Setup	/cc-best:setup, /cc-best:setup-pm, /cc-best:status, /cc-best:self-check, /cc-best:confidence-check, /cc-best:security-audit	Configuration & diagnostics

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

---

🛠️ Skills

19 development skills organized by domain:

Domain	Skills	Coverage
Backend	backend, api, database	Python, TS, Java, Go, C#
Frontend	frontend, native	Web + iOS/macOS/Tauri
Quality	quality, testing, security, debug	TDD, OWASP, profiling
Architecture	architecture, devops, git	ADR, CI/CD, branching
Routing	model	Task→model recommendation
Session	session, learning, compact, exploration	Lifecycle + knowledge management
Research	search-first, second-opinion	Search strategy, cross-validation

📖 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 (30)"]
        H1["PreToolUse: validate, secrets, protect"]
        H2["PostToolUse: format, typecheck, observe"]
    end

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

Layer	Trigger	Purpose
Commands	User types /xxx	Role workflow, user-initiated actions
Skills	Auto-injected	Best practices, coding standards
Agents	Task tool delegation	Specialized sub-tasks (review, planning)
Hooks	Lifecycle events	Safety guards, auto-format, pattern learning

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

CC-Best Content	Official Plugin	Relationship
code-reviewer agent	code-review plugin	Built-in: lightweight local version; Official: more powerful with auto-trigger
security-reviewer agent	security-guidance	Built-in: OWASP checklist; Official: automatic security analysis
code-simplifier agent	code-simplifier	Similar function; official plugin has more context
/cc-best:cc-ralph command	ralph-loop plugin	CC-Best wrapper; requires plugin for cross-session persistence
hookify examples	hookify plugin	Built-in: examples; Official: full hook management

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/.

Trigger	Function	Script (Node.js)
PreToolUse	Validate dangerous commands	validate-command.js
PreToolUse	Confirm before git push	pause-before-push.js
PreToolUse	Protect sensitive files	protect-files.js
PreToolUse	Block random .md creation	block-random-md.js
PreToolUse	Long-running task warning	long-running-warning.js
PostToolUse	Auto-format code	format-file.js
PostToolUse	Check console.log	check-console-log.js
PostToolUse	TypeScript type check	typescript-check.js
SessionStart	Session health check	session-check.js
PreCompact	Save state before compact	pre-compact.js

---

💡 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:

Directory	Source	Purpose
.playwright-mcp/	MCP auto-created	Playwright MCP temporary files
.claude/mcp-data/	MCP auto-created	MCP shared data
*-mcp/	MCP auto-created	Other MCP tool directories
.claude/screenshots/	Template-defined	Manually saved screenshots (meaningful)

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

Mode	Control	Use Case
/cc-best:iterate	Fully autonomous	Clear task list
/cc-best:pair	Confirm each step	Learning, sensitive ops

• 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:

Scenario	Recommended	Why
Team collaboration	CC-Best	Role workflow (PM→Lead→Dev→QA)
Multi-language stack	CC-Best	7 language coding standard dirs
Chinese team	CC-Best	Bilingual docs
Solo developer	Superpowers	Lighter, git worktree automation
Need git worktree	Superpowers	Auto-creates isolated branches

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

---

🔧 Requirements

Dependency	Version	Notes
Claude Code	Latest recommended	Hooks require recent versions
Node.js	16+	For cross-platform hooks (default)
Python	3.8+	For some hook scripts
Bash/Git Bash	Any version	Optional for bash hooks

Optional MCP Servers

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

MCP Server	Used By	Purpose
Playwright	/cc-best:designer, /cc-best:dev, /cc-best:pm	Browser automation for UI testing and screenshots
Firecrawl	/cc-best:pm, /cc-best:lead	Web scraping for requirement research

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

Supported Languages

Language	Rule File	Formatter	Test Framework
Python	python-style.md	Black + isort	pytest
Vue/TS/JS	frontend-style.md	Prettier	Vitest
C++	cpp-style.md	clang-format	Google Test
Java	java-style.md	google-java-format	JUnit
C#	csharp-style.md	dotnet format	xUnit/NUnit
Go	backend/go.md	gofmt	testing
Swift	native/ios.md	swift-format	XCTest

---

📚 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.

Contribution Type	Description
⭐ Star	Show your support
🐛 Bug Report	Report issues
💡 Feature Request	Suggest features
📝 Documentation	Improve docs
🔧 Code	Add commands, rules, skills

---

📄 License

MIT License - Free to use and modify

---

Star History