feat: Add research-backed model-specific prompt optimization

- Implement flexible LLM detection (OpenAI/GPT/ChatGPT → GPT-4, etc.)
- Add philosophy-driven optimization based on official documentation
- Integrate advanced techniques: agentic control, XML structure, PTCF framework
- Add real-time quality scoring and production-ready templates
- Create comprehensive documentation suite with comparative analysis

Based on deep research of OpenAI, Anthropic, and Google prompting methodologies.

Author: gauravsurtani

.kiro/steering/prompting-best-practices.md

@@ -0,0 +1,76 @@
+# Prompting Best Practices Knowledge Base
+
+## Master Prompt Generation Rules
+
+### Universal Structure Requirements
+All generated prompts MUST include these components:
+- [ROLE]: Specific persona or expertise level
+- [DIRECTIVE]: Clear, actionable instruction using specific verbs
+- [CONTEXT]: Essential background information and constraints
+- [OUTPUT_FORMAT]: Exact structure, length, and formatting requirements
+- [TONE]: Communication style and target audience
+- [CONSTRAINTS]: Limitations, rules, and what to avoid
+- [EXAMPLES]: Input-output demonstrations (2-3 high-quality examples)
+- [CREATIVITY_LEVEL]: Factual precision vs. creative exploration guidance
+- [ERROR_HANDLING]: Instructions for uncertainty and missing information
+
+### Model-Specific Optimization Rules
+
+#### OpenAI GPT-4 - "Configurable Agent" Philosophy
+**Source**: https://platform.openai.com/docs/guides/prompt-engineering
+- Place instructions at beginning with delimiters (###, """)
+- Use positive framing - say what TO do, not what NOT to do
+- Implement task decomposition for complex problems
+- Add <self_reflection> tags for agentic control
+- Configure reasoning_effort (low/medium/high) for computational budget
+- Separate instructions from context with clear structural boundaries
+- Temperature: 0.0-0.3 factual, 0.7-1.0 creative, 1.0-2.0 experimental
+
+#### Anthropic Claude - "Auditable Processor" Philosophy  
+**Source**: https://docs.anthropic.com/claude/docs/prompt-engineering
+- Use semantic XML tags: <instructions>, <thinking>, <answer>, <document>
+- Be explicit and direct - avoid all ambiguity
+- Place long documents at TOP, query at END for attention optimization
+- Use response prefilling (start with "{") to force JSON format
+- Implement prompt chaining for complex multi-step workflows
+- Mandatory <thinking> tags for transparent, auditable reasoning
+- Apply Constitutional AI principles for safety-first approach
+
+#### Google Gemini - "Creative Collaborator" Philosophy
+**Source**: https://ai.google.dev/gemini-api/docs/prompting-strategies  
+- Use PTCF framework: Persona, Task, Context, Format
+- Make it conversational - treat as collaborative partner
+- Leverage native multimodality (text + images + audio)
+- Use iterative refinement through follow-up prompts
+- Describe scenes narratively, don't just list keywords
+- Configure Temperature/topK/topP for creative control
+- Start good, refine through conversational iteration
+
+### Quality Assessment Criteria
+- Role Definition (10 points): Clear, specific persona
+- Directive Clarity (15 points): Unambiguous, actionable instruction
+- Context Completeness (10 points): Sufficient background information
+- Output Format (10 points): Specific structure requirements
+- Examples Quality (15 points): Relevant, diverse demonstrations
+- Constraints (10 points): Clear limitations and rules
+- Error Handling (10 points): Uncertainty management
+- Tone Specification (5 points): Appropriate communication style
+- Creativity Guidance (5 points): Factual vs. creative balance
+- Overall Coherence (10 points): Logical flow and completeness
+
+### Safety and Ethics Requirements
+- Include bias prevention measures
+- Specify appropriate content boundaries
+- Add harm prevention guidelines
+- Request fact-checking for factual claims
+- Use inclusive language and examples
+- Respect privacy and confidentiality
+
+### Enhancement Process Rules
+When enhancing prompts:
+1. Preserve ALL original requirements and constraints
+2. Apply model-specific optimization techniques from official sources
+3. Ensure production-ready quality and reliability (80%+ quality score)
+4. Include appropriate safety and quality controls
+5. Use model-preferred structure and formatting
+6. Validate that enhancement improves clarity without changing intent

.vscode/settings.json

@@ -0,0 +1,2 @@
+{
+}

README.md

@@ -1,28 +1,56 @@
 # Master Prompt Creator
 
-A web-based tool designed to help users construct high-quality, well-structured, and effective prompts for Large Language Models (LLMs) through a guided, step-by-step process.
+A web-based tool designed to help users construct high-quality, well-structured, and effective prompts for Large Language Models (LLMs) through a guided, step-by-step process with model-specific optimization.
 
 ## Overview
 
-The Master Prompt Creator transforms a user's basic idea into a detailed, optimized prompt. It employs a wizard-like interface to ask clarifying questions based on established prompt engineering principles. The final output provides both a raw, structured prompt and an AI-enhanced version, leveraging the Gemini API for automated optimization.
-
-## Key Features
-
--   **Guided Prompt Construction**: A multi-step questionnaire guides users through defining the core components of a good prompt: Role, Directive, Context, Constraints, Output Format, Tone, and Examples.
--   **AI-Powered Example Generation**: If a user needs help creating few-shot examples, the tool can call the Gemini API to generate relevant input/output pairs based on the task description.
--   **Automated Prompt Enhancement**: After the user builds their "raw" prompt, the tool makes a second API call to have an "expert prompt engineer" (simulated by another LLM) refine and optimize the prompt for clarity, conciseness, and effectiveness.
--   **Side-by-Side Comparison**: The final screen displays the user-generated raw prompt and the AI-enhanced version, allowing for easy comparison and copying.
--   **Zero Dependencies**: The entire application is a single `index.html` file with no external frameworks, making it extremely portable and easy to run.
--   **Clean, Responsive UI**: Styled with Tailwind CSS for a modern and intuitive user experience on all devices.
-
-## How It Works
-
-1.  **Initial Task**: The user starts by entering the core task they want the LLM to perform.
-2.  **Clarification Wizard**: The application presents a series of questions, one by one. The user's answers are saved as they progress through the steps.
-3.  **Generate Examples (Optional)**: On the "Examples" step, the user can click a button to have the Gemini API generate examples for them.
-4.  **Generate Master Prompt**: Once all questions are answered, the tool assembles the answers into a structured "Raw Prompt."
-5.  **Enhance with AI**: Concurrently, the raw prompt is sent to the Gemini API with a meta-prompt asking it to act as an expert and improve the prompt.
-6.  **Display & Use**: Both prompts are displayed. The user can copy their preferred version with a single click.
+The Master Prompt Creator transforms a user's basic idea into a detailed, optimized prompt using official best practices from OpenAI, Anthropic, and Google. It employs a wizard-like interface to ask clarifying questions based on established prompt engineering principles, then generates both a structured prompt and an AI-enhanced version optimized for the user's target LLM.
+
+## 🎯 Key Features
+
+### **Core Functionality**
+-   **Guided Prompt Construction**: Multi-step questionnaire covering Role, Directive, Context, Constraints, Output Format, Tone, Examples, Creativity Level, and Error Handling
+-   **Model-Specific Optimization**: Tailored enhancement based on target LLM (GPT-4, Claude, Gemini)
+-   **AI-Powered Example Generation**: Gemini API generates relevant input-output pairs
+-   **Automated Prompt Enhancement**: Expert-level prompt optimization using model-specific best practices
+-   **Quality Assessment**: Real-time prompt scoring (0-100%) with improvement recommendations
+
+### **Advanced Features**
+-   **Official Resource Integration**: Direct links to OpenAI, Anthropic, and Google prompting guides
+-   **Model-Specific Techniques**: Applies platform-specific optimization (XML tags for Claude, system messages for GPT-4, etc.)
+-   **Quality Indicators**: Visual scoring with green/yellow/red quality badges
+-   **Side-by-Side Comparison**: Raw vs. enhanced prompts with copy functionality
+-   **Zero Dependencies**: Single HTML file with no external frameworks
+-   **Responsive Design**: Modern UI with Tailwind CSS
+
+## 🚀 How It Works
+
+### **Step-by-Step Process**
+1.  **Initial Task**: User enters the core task they want the LLM to perform
+2.  **Guided Questionnaire**: 8-step wizard covering all essential prompt components:
+    - Directive (specific action/command)
+    - Role/Persona (expert identity)
+    - Context (background information)
+    - Tone & Audience (communication style)
+    - Output Format (structure requirements)
+    - Constraints (limitations/rules)
+    - Examples (few-shot demonstrations)
+    - Creativity Level (factual vs. creative approach)
+    - Error Handling (uncertainty management)
+    - Target LLM (optimization preference)
+
+3.  **AI-Powered Assistance**: 
+    - Generate examples automatically using Gemini API
+    - Refine text with AI enhancement at any step
+
+4.  **Quality Assessment**: Real-time prompt scoring with specific recommendations
+
+5.  **Model-Specific Enhancement**: 
+    - GPT-4: System messages, few-shot prompting, chain-of-thought
+    - Claude: XML structure, Constitutional AI principles
+    - Gemini: System instructions, structured output, multimodal support
+
+6.  **Final Output**: Side-by-side comparison of raw and enhanced prompts with quality indicators
 
 ## Technology Stack
 
@@ -53,6 +81,52 @@ This application is designed to be secure and does not come with a pre-loaded AP
 
 Your API key is saved securely in your browser's `localStorage` and is **never shared or stored outside of your machine**. You only need to do this once per browser.
 
+## 📚 Documentation
+
+### **Core Documentation**
+- **[Master Prompt Generation Rules](master-prompt-generation-rules.md)**: Comprehensive rules and structure for generating high-quality prompts
+- **[Official Best Practices Summary](official-prompting-best-practices-summary.md)**: Key principles from OpenAI, Anthropic, and Google documentation
+- **[Official Resources](official-prompting-resources.md)**: Direct links to all official prompting guides and examples
+
+### **Technical Guides**
+- **[Comprehensive Guidelines](prompting-guidelines-comprehensive.md)**: Detailed prompting techniques and strategies
+- **[Steering Rules](.kiro/steering/prompting-best-practices.md)**: Active rules used by the application for prompt generation
+
+### **Model-Specific Optimization**
+The application automatically applies best practices from:
+- **OpenAI GPT-4**: https://platform.openai.com/docs/guides/prompt-engineering
+- **Anthropic Claude**: https://docs.anthropic.com/claude/docs/prompt-engineering  
+- **Google Gemini**: https://ai.google.dev/gemini-api/docs/prompting-strategies
+
+### **Quality Standards**
+- **80%+ Quality Score**: Production-ready prompts with comprehensive components
+- **Model-Specific Techniques**: Automatic application of platform-optimized structures
+- **Safety Guidelines**: Built-in bias prevention and content safety measures
+- **Official Compliance**: All techniques sourced from official documentation
+
+## 🎯 Usage Examples
+
+### **For GPT-4 Optimization**
+Set Target LLM to "GPT-4" to automatically apply:
+- System message structure
+- Few-shot prompting with examples
+- Chain-of-thought reasoning
+- Temperature control guidance
+
+### **For Claude Optimization**  
+Set Target LLM to "Claude" to automatically apply:
+- XML tag structure (`<thinking>`, `<context>`, `<output>`)
+- Constitutional AI principles
+- Direct, explicit instructions
+- Long context utilization
+
+### **For Gemini Optimization**
+Set Target LLM to "Gemini" to automatically apply:
+- System instruction format
+- Structured output formatting
+- Multimodal considerations
+- Safety settings integration
+
 ## License
 
 This project is licensed under the MIT License - see the [LICENSE](LICENSE) file for details.