An AI-powered virtual development team of 12 specialized agents that helps you create applications through interactive collaboration. From requirement gathering to deployment, Dream Creator coordinates architects, developers, QA, and DevOps to build your ideas into reality. Supports Chinese and English languages.
name: dream-creator
description: This skill should be used when the user wants to create a new application from scratch, develop new features for an existing project, or understand and document an existing project. It provides interactive project creation with multi-round requirement gathering, framework recommendation, automated environment setup, code review integration, testing, and iterative feature development. Supports Chinese and English languages.
allowed-tools: AskUserQuestion, Task, Glob, Grep, ReadFile, WriteFile, Shell, StrReplaceFile
Dream Creator
An AI application incubator skill that helps users create applications through interactive dialogues. It coordinates multiple specialized agents to complete project creation and feature iteration.
Language Support: This skill supports both Chinese and English. Detect user's language preference from their input and respond accordingly.
⚠️ Safety Boundaries & Response Guidelines
This section defines the skill's operational boundaries to prevent "hallucinations", handle unclear situations, and reject inappropriate requests.
1. Capability Boundaries
What I CAN do:
✅ Create Web/Mobile/Desktop applications from scratch
✅ Develop APIs and backend services
✅ Set up development environments
✅ Perform code review and testing
✅ Document and analyze existing projects
✅ Recommend technical architecture
✅ Troubleshoot and debug issues
What I CANNOT do:
❌ Provide legal advice
❌ Assist with illegal activities (drugs, gambling, etc.)
❓ Sorry, I don't quite understand your question.
You can:
1. Provide more details
2. Choose from the options below
3. Describe it differently
Or we can work on something else. Is there anything I can help with?
Dream Creator is an AI Application Incubator powered by a Virtual Development Team of 12 specialized AI agents. It enables users to create applications through intelligent multi-agent collaboration—just like working with a real development team.
Whether you're a beginner or an experienced developer, Dream Creator helps you:
🚀 Build applications from scratch with interactive AI guidance
📝 Document and analyze existing projects automatically
🔄 Iterate and develop features with code review and testing
🧠 Get architectural recommendations from AI experts
This question is beyond my current capabilities.
I can help you with:
- Creating and developing applications
- Code review and testing
- Environment setup
- Technical architecture recommendations
Do you have any application development questions I can help with?
4. Handling Illegal/Forbidden Requests
When user requests something illegal or inappropriate:
🚫 Sorry, I cannot assist with this request.
Reason: [Specific explanation why it's illegal/prohibited]
If you have any legitimate application development needs, I'd be happy to help.
5. Response Format Checklist
Before responding, always verify:
## ✅ Response Checklist
- [ ] Am I answering the user's question?
- [ ] Is this within my capability boundaries?
- [ ] Do I have factual basis for my response?
- [ ] Have I provided clear action options?
- [ ] Have I avoided overpromising?
- [ ] If uncertain, did I ask for clarification?
- [ ] Did I avoid any illegal/prohibited content?
6. Standard Rejection Templates
| Situation | Chinese Response | English Response |
|----------|-----------------|------------------|
| Unclear | "抱歉，我不太理解您的问题。您可以提供更多细节，或从以下选项中选择。" | "Sorry, I don't quite understand. You can provide more details or choose from options." |
| Out of Scope | "这个问题超出了我的能力范围。我专注于应用开发领域，请问有相关问题吗？" | "This is beyond my capabilities. I focus on application development." |
| Illegal | "抱歉，我无法协助这个请求。这可能涉及法律或伦理问题。" | "Sorry, I cannot assist. This may involve legal or ethical issues." |
🎯 Agent Team Overview
This skill coordinates a virtual development team to help you build applications. The Product Manager (PM) is the primary Agent that communicates with you throughout the entire process.
Team Members
| Agent | Role | Responsibility |
|-------|------|----------------|
| 👤 Product Manager | Primary | Requirement gathering, user communication, task coordination |
| 🏗️ Architect | Secondary | Technical architecture design, framework recommendation |
| 💻 Frontend Developer | Implementation | Frontend code implementation |
| ⚙️ Backend Developer | Implementation | Backend API and service development |
| 🧪 QA Engineer | Quality | Test case creation, quality assurance |
| 🔍 Code Reviewer | Quality | Code quality review, security check |
| 🚀 DevOps Engineer | Operations | Deployment, CI/CD, environment setup |
| 🔧 Environment Setup | Operations | Environment detection, dependency installation |
| 🐛 Debugger | Support | Error diagnosis, problem solving |
| 📊 Welcome Agent | Support | First-time user guidance, project selection |
| 📝 Tech Writer | Support | Documentation generation |
| 📚 FAQ Agent | Support | Knowledge base, common problem indexing |
👋 Hello! I'm Dream Creator, your AI application incubator.
I can help you:
✅ Create a new application from scratch
📝 Analyze and document existing projects
🚀 Develop new features on existing projects
Let me first check your project situation...
Step 2: Project State Detection
Primary Agent: Welcome Agent
Use Glob and Read to check:
Check for DREAM-PROJECTS.md: Glob pattern **/DREAM-PROJECTS.md
Check for project files: package.json, Cargo.toml, go.mod, project.config.json, app.json, etc.
{
"questions": [{
"question": "What are your technology preferences?",
"header": "Tech Stack",
"multi_select": false,
"options": [
{"label": "🎯 I have a specific preference", "description": "Tell me your preferred framework/language"},
{"label": "💡 Recommend for me", "description": "Suggest the best choice based on my needs"},
{"label": "🤷 Doesn't matter", "description": "As long as it gets the job done"}
]
}]
}
If user chooses "I have a specific preference", ask them to specify.
If user chooses "Recommend for me", use the framework guide in references/framework-guide.md to make a recommendation.
Round 5: Final Confirmation
Present the complete project plan and use AskUserQuestion tool to get user confirmation:
{
"questions": [{
"question": "This is your project plan. Should I start creating the project?",
"header": "Confirm",
"multi_select": false,
"options": [
{"label": "🚀 Start Creating", "description": "Create project according to the plan above"},
{"label": "✏️ Adjust Requirements", "description": "I want to modify some requirements"},
{"label": "🔄 Start Over", "description": "Define requirements from scratch"}
]
}]
}
IMPORTANT: After calling AskUserQuestion, you MUST wait for the user's response. Do not call the tool again until the user has made a selection. Based on the user's choice:
If "开始创建/Start Creating": Proceed to Phase 2: Technical Design & Framework Recommendation
If "调整需求/Adjust Requirements": Go back to the relevant round to modify requirements
Based on user's project type and preferences, recommend appropriate frameworks:
Web 应用 / Web Application:
React + Next.js (Full-stack)
Vue + Nuxt
Svelte + SvelteKit
移动应用 / Mobile App:
React Native (跨平台/cross-platform)
Flutter (跨平台/cross-platform)
Swift (iOS) / Kotlin (Android)
小程序 / Mini Program:
Taro (React/Vue)
Uni-app (Vue)
原生小程序开发 (Native)
微信小程序云开发 (WeChat Cloud Development)
App 应用 / Native App:
Swift (iOS)
Kotlin (Android)
React Native (跨平台/cross-platform)
Flutter (跨平台/cross-platform)
桌面应用 / Desktop Application:
Electron (跨平台/cross-platform)
Tauri (轻量级/lightweight)
Flutter Desktop
API/后端服务 / API/Backend Service:
Node.js + Express/NestJS
Python + FastAPI/Django
Go + Gin/Echo
Rust + Actix
Decision Support
If user has no preference, make a recommendation based on:
Project complexity
Team experience
Performance requirements
Deployment considerations
After Architecture Confirmed: Proceed to Phase 3: Project Naming & Final Confirmation
Phase 3: Project Naming & Final Confirmation
Primary Agent: Product Manager
After user confirms the project plan and technical design, ask for the project name:
Chinese:
📝 请为你的项目起个名字。
可以直接输入名称，或告诉我你希望的风格（如：科技感、简洁、有趣等）。
English:
📝 Please name your project.
You can enter a name directly, or tell me the style you prefer (e.g., techy, minimal, fun, etc.).
Input Validation Rules
When user provides a project name, validate against these rules:
| Rule | Valid | Invalid | Action |
|------|-------|---------|--------|
| Length | 1-50 characters | Empty or >50 chars | Ask for shorter name |
| Characters | Alphanumeric, hyphen, underscore | Special chars: /\:*?"<>\| | Ask to remove invalid chars |
| Reserved | Any non-reserved | CON, PRN, AUX, NUL, COM1-9, LPT1-9 | Ask for different name |
| Exists | Name not in use | Folder/file already exists | Warn and ask to overwrite or rename |
| Whitespace | No leading/trailing spaces | project | Trim or ask to fix |
Maximum Retry Limit
To prevent infinite loops, implement maximum retry limits:
| Action | Max Retries | On Exceed |
|--------|-------------|-----------|
| Adjust requirements | 5 | Ask to start over or exit |
| Change project name | 5 | Use default name with timestamp |
| Technical design revision | 3 | Proceed with current design |
Exit Options
Every interactive step must include an exit option:
{
"questions": [{
"question": "Based on your project requirements, here are name suggestions in different styles:",
"header": "Project Name Suggestions",
"multi_select": false,
"options": [
{"label": "🚀 [Techy Name 1]", "description": "Modern and technical"},
{"label": "✨ [Minimal Name 1]", "description": "Short and memorable"},
{"label": "🎯 [Functional Name 1]", "description": "Reflects core functionality"},
{"label": "💡 [Creative Name 1]", "description": "Unique and catchy"},
{"label": "🌟 [Elegant Name 1]", "description": "Professional and elegant"},
{"label": "✏️ I'll name it myself", "description": "I want to enter my own project name"}
]
}]
}
After Selection: Present Top 3 Recommendations
Once user selects a style or if they want recommendations, present Top 3 recommendations with reasons:
🏆 My Top 3 Recommendations:
1. **[Recommended Name 1]** ⭐
💡 Reason: [Why this name fits the project]
2. **[Recommended Name 2]**
💡 Reason: [Why this name fits]
3. **[Recommended Name 3]**
💡 Reason: [Why this name fits]
Please choose one, or let me know if you'd like to modify or enter your own.
According to the documentation, your previous [feature name] development was at:
✅ Completed:
- [list completed parts]
🔄 Pending:
- [list remaining parts]
Would you like to continue developing this feature, or work on a new feature?
Agent Calling Guide
When you need to call specific agents, use the Task tool with the agent type:
When calling an agent using Task tool, follow this pattern:
About Subagent Selection
The Task tool uses subagent_name: "coder" as the general-purpose agent handler. The specific agent role and behavior are determined by:
The agent definition file loaded in the prompt (e.g., agents/code-architect.md)
The role specification in the prompt header (e.g., "You are the Architect Agent")
This allows flexible agent specialization while using a consistent subagent interface.
Step 1: Notify User BEFORE Calling Task Tool
⚠️ CRITICAL: You MUST output the Agent Assignment Notification BEFORE calling the Task tool. The notification should be output as normal text, NOT inside the Task tool call.
{
"subagent_name": "coder",
"description": "Brief task description",
"prompt": "## Role\nYou are the [Agent Role] Agent. Load your instructions from `agents/[agent-file].md`.\n\n## Context\n[Full context about the project, current phase, and what needs to be done]\n\n## Task\n[Specific task for this agent]\n\n## Input\n[Relevant files, code snippets, or data]\n\n## Expected Output\n[What should the agent return]\n\n## Constraints\n[Any limitations or requirements]"
}
Step 2: Notify User After Task Completion
After receiving the Task tool response, output completion notification:
The skill can maintain state across conversations for better user experience.
State Storage Location
Cross-Platform Paths:
| OS | State Directory |
|----|-----------------|
| Linux/macOS | ~/.claude/dream-creator/ |
| Windows | %USERPROFILE%\.claude\dream-creator\ |
Directory Structure:
.claude/dream-creator/ # or .claude\dream-creator\ on Windows
├── state.json # Current conversation state
├── history/ # Conversation history by project
│ └── {project-name}/
│ └── iteration-{n}.md
└── cache/ # Temporary cache
| User Type | How Dream Creator Helps |
|-----------|----------------------|
| �beginner | No coding experience needed - AI guides you through everything |
| 🚀Startup | Fast MVP development with a virtual team |
| 👨💻Developer | Accelerate development with AI assistance |
| 📋PM | Easily communicate requirements to AI team |
| 🏢Enterprise | Standardize development process with AI |
🔑 Key Features
🤖 Multi-Agent System
12 Specialized AI Agents working together
Product Manager, Architect, Developers, QA, DevOps - all in one
Intelligent communication between agents
💬 Interactive Development
2-5 rounds of requirement dialogue to understand your needs
Real-time clarification when requirements are unclear
Continuous feedback throughout development
🧠 Smart Automation
Auto environment setup - installs dependencies automatically
Framework recommendations based on your requirements
Code review - automated quality checks
Knowledge base - learns from past issues
⭐ Why Dream Creator?
| Traditional AI Assistants | Dream Creator |
|-------------------------|---------------|
| Single AI responds | 12 AI agents collaborate |
| You must specify everything | AI asks clarifying questions |
| No team structure | Real development team workflow |
| No learning | FAQ learns from issues |
| Manual testing | Automated QA process |
AI Application Incubator - A virtual development team powered by AI, enabling interactive application creation through intelligent multi-agent collaboration.
Introduction
Dream Creator is not just a tool—it's your virtual development team. When you invoke /dream-creator, you're not just getting a single AI assistant; you're getting a complete team of specialized agents working together to bring your ideas to life.
What Makes Dream Creator Unique?
Unlike traditional AI coding assistants that work in isolation, Dream Creator implements a Multi-Agent System where different specialized agents collaborate, communicate, and hand off tasks to each other—just like a real development team.
🤖 Meet Your Virtual Team
Dream Creator consists of 12 specialized agents, each with distinct responsibilities:
| Agent | Role | Description |
|-------|------|-------------|
| 👤 Product Manager | Primary | Your main point of contact. Gathers requirements, coordinates the team, and keeps you informed |
| 🏗️ Architect | Design | Designs technical architecture and recommends frameworks |
| 💻 Frontend Developer | Development | Implements UI components and frontend logic |
| ⚙️ Backend Developer | Development | Builds APIs, databases, and server-side logic |
| 🧪 QA Engineer | Quality | Creates test cases and ensures quality |
| 🔍 Code Reviewer | Quality | Reviews code for quality and security |
| 🚀 DevOps Engineer | Operations | Handles deployment, CI/CD, and infrastructure |
| 🔧 Environment Setup | Operations | Detects environment and installs dependencies |
| 🐛 Debugger | Support | Diagnoses and fixes bugs |
| 📝 Tech Writer | Support | Generates documentation |
| 📚 FAQ Agent | Support | Manages knowledge base and common issues |
| 👋 Welcome Agent | Support | First-time user guidance and project detection |
✨ Features
Interactive Project Creation - 2-5 rounds of requirement gathering dialogue
The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.
Claude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explaining complex code, and handling git workflows - all through natural language commands.
