Prompt Gallery: The MVC Prompt Framework

📖 Introduction

Stop writing spaghetti prompts!

As developers, we wouldn't write an entire application in a single main() function. Yet, most LLM prompts are unstructured walls of text that mix persona, logic, data, and formatting instructions.

Prompt Gallery is an open repository dedicated to standardizing Prompt Engineering via the Model-View-Controller (MVC) pattern.

I maintain a strict markup-based framework that separates concerns:

• Model: The knowledge, state schema, and reasoning logic.
• View: The output formatting and presentation layer.
• Controller: The session flow, user interaction rules, and guardrails.

🏗 The Architecture

This repository uses a <MASTER_PROMPT> template the [prompt-template.md](prompt-template.md) file. Every use case in this gallery must adhere to this structure.

Why MVC for LLMs?

1. Maintainability: You can tweak the View (output format) without breaking the Controller (logic).
2. Predictability: By defining a <STATE_SCHEMA>, we force the LLM to maintain consistency across long sessions.
3. Scalability: Complex agent behaviors become manageable when broken down into components.

The Template Structure

Every prompt in this gallery is built on the following skeleton:

<MASTER_PROMPT>
<!-- 
    MASTER_PROMPT: 
    The root container for all configuration layers. 
    Serves as the complete specification for transforming the LLM into an MVC-style system.
-->
    <CORE_DIRECTIVES>
        <!-- 
            CORE_DIRECTIVES:
            Defines identity, constraints, purpose, and non-negotiable rules.
            This section overrides all others and sets the "constitution" of the system.
        -->
        <PERSONA>
            <!-- 
                PERSONA:
                Defines how the LLM should behave and sound. 
                Should not include task-specific instructions, only character and style.
            -->
            <ROLE></ROLE>
            <TONE_OF_VOICE>
                <COMMUNICATION_STYLE></COMMUNICATION_STYLE>
            </TONE_OF_VOICE>
        </PERSONA>
        <VISION>
            <!-- 
                VISION:
                The long-term philosophical intent behind the system.
                Example: "Deliver structured, predictable, modular outputs." 
            -->
        </VISION>
        <MISSION>
            <!-- 
                MISSION:
                Describes the operational goal for this session or agent.
                Example: "Help the user achieve X by applying MVC logic."
            -->
        </MISSION>
        <ABSOLUTE_RULES></ABSOLUTE_RULES>
    </CORE_DIRECTIVES>

    <IN_PROMPT_CONTEXT>
        <!-- 
            IN_PROMPT_CONTEXT:
            Contains all situational/background info provided by the user or system.
            Should not include directives. Only facts, history, or input data.
        -->
    </IN_PROMPT_CONTEXT>

    <MODEL>
        <DIRECTIVES></DIRECTIVES>
        <STATE_SCHEMA>
            <!-- 
                STATE_SCHEMA:
                The actual data, schemas, or knowledge references used by the model.
                May include static definitions or dynamic memory for the session.
            -->
        </STATE_SCHEMA> 
    </MODEL>

    <VIEW>
         <!-- 
            VIEW:
            Defines how content is formatted, structured, and delivered to the user.
            Only concerns presentation—no logic or reasoning policies.
        -->
        <DIRECTIVES></DIRECTIVES>
        <TEMPLATES></TEMPLATES>
    </VIEW>

    <CONTROLLER>
        <!-- 
            CONTROLLER:
            Manages interaction flow, handles user input, 
            decides which part of the model/view to call, and enforces session logic.
        -->
        <DIRECTIVES></DIRECTIVES>
        <SESSION_PHASES>
            <!-- 
                SESSION_PHASES:
                Instructions for iterative cycles (e.g., observe → think → respond → request input).
                Defines when and how the controller loops back to earlier phases.
            -->
        </SESSION_PHASES>
        <SESSION_LOOP>
            <!-- 
                SESSION_LOOP:
                Instructions for iterative cycles (e.g., observe → think → respond → request input).
                Defines when and how the controller loops back to earlier phases.
            -->
        </SESSION_LOOP>
        <SESSION_RULES>
            <!-- 
                SESSION_RULES:
                Guardrails for managing user messages, interruptions, clarifications, error states, and transitions.
            -->
        </SESSION_RULES>
    </CONTROLLER>

</MASTER_PROMPT>

🚀 Getting Started

1. Clone the repository:

   git clone [https://github.com/jvanheerikhuize/prompt-gallery.git](https://github.com/jvanheerikhuize/prompt-gallery.git)

2. Pick a use-case: Browse the /Fun or /Productivity directory.
3. Deploy: Copy the raw content of the prompt-template.md file and paste it into your LLM of choice (Claude, GPT-4, etc.).

🤝 How to Contribute

We are looking for robust, production-ready configurations.

I need use cases for:

• Drawing in ASCII: A prompt that acts as a text based designer.
• Performance Abstraction: A prompt that takes a prompt and compresses it for performance without quality loss.
• LLM Benchmarker: A prompt that is able to scan its'model for features.

Contribution Steps:

1. Fork the repository.
2. Create a new file in the appropriate folder or add one.
3. Implement the MVC tags using the prompt-template.md as a base.
4. Test your prompt to ensure the Controller enforces the rules defined in the Model.
5. Submit a Pull Request.

📄 License

This project is licensed under the Apache 2.0 License - see the LICENSE file for details.