A modular, customizable framework for building Prompt-Driven AI Assistants — where behavior, tone, and structure are fully defined through prompt layers.
This template helps you design powerful, strategic, and creative assistants using a transparent, modular prompt system — not hidden code or black-box logic.
It supports role-based behavior, drift-level creativity, bias control, and collaborative workflows.
Follow these steps to get your assistant running:
- Open
template-GPT.mdand define your assistant’s name, purpose, and mission. - Select your roles, lenses, drift level, output modes, and bias/fact-check mode.
- Review the output structure and align examples and formats.
- Choose your platform (OpenAI, Claude, etc.) and follow platform-specific setup.
- Use the Customization Checklist and Refinement Loop for iteration.
- Test, deploy, and adapt your assistant.
- Instructions Field Limit: 8,000 characters (including spaces and line breaks).
- Workarounds:
- Move non-essential or reusable content (detailed examples, presets, configuration notes) to the “Knowledge” section or to attached files.
- Use concise, modular instructions in the main field, referencing external content for details.
- For large templates: Split instructions into logical modules and reference them in the main prompt, or use a “prompt chaining” approach. Prompt chaining means breaking complex tasks into sequential steps, where each step’s output informs the next.
- Best Practices:
- Define WHO, WHAT, and WHY at the start of your instructions.
- Break down complex instructions into clear steps, using delimiters and few-shot examples where needed.
- Specify output format and style for consistency.
- Use Markdown formatting for improved readability.
- Prompt Structure: Claude responds well to clear, structured prompts with explicit instructions and examples.
- Best Practices:
- Use XML or Markdown to separate instructions, context, and examples.
- Assign a clear role at the start (“You are a [role]...”) and specify the expected output format.
- Prefill the start of the assistant’s message to control output structure and tone.
- Use prompt chaining or step-by-step instructions for complex tasks.
- Claude’s context window is large (up to 200,000 tokens), but keep prompts concise for clarity.
- Adaptation Tips:
- Remove or condense OpenAI-specific references.
- Adjust formatting if using XML tags for Claude.
- Test and iterate to optimize for Claude’s conversational and analytical strengths.
For other LLMs (e.g., Gemini, Mistral): Adapt the prompt structure as above, using clear instructions and examples. Test and adjust for each model’s strengths and quirks.
- Modular design with roles, lenses, and personality modes.
- Interpretive drift levels from Literal to Unmoored.
- Output intent + flavor for tone and use case control.
- GEO-aware assistant design for AI citation, discoverability, and structured formatting.
- Built-in Bias & Fact-Check Mode (Strict / Flexible / Creative).
- Structured outputs and behavioral principles.
- Presets for common tasks + example prompts.
- Refinement loop for iterative feedback.
- Testing, validation, troubleshooting, and collaborative tools.
- Optional PrecisionMode for stricter factual control (Strict / Adaptive / Experimental).
- Supports prompt engineering techniques like Emotion Shaping or Scrambled Prompting.
- Prompt Sharpening Guide for GPT‑4.1, Claude, Gemini (role clarity, micro-steps, scannable structure)
- Platform-aware formatting logic for cinematic or generative tools (Runway, Pika, etc.)
- Optional multimodal input handling for sketches, sound, or reference video
- Preset Modes for cinematic/symbolic assistants (e.g., Visual Script, Symbolic Drift)
| Section | Description |
|---|---|
| Purpose | Defines the assistant’s core objectives and value proposition. |
| User Input Guidelines | Details accepted input types, best practices, and constraints. |
| Prompt Sharpening Guide | How to write clear, modular, example-driven prompts for GPT-4.1, Claude, Gemini. |
| Roles | Behavioral perspectives (e.g., Strategic Partner, Creative Director). |
| Lenses | Interpretive contexts (e.g., Narrative Layering, Strategic Focus). |
| Drift Levels | Range from literal to recursive interpretation. |
| Output Modes | Intent (Concept, Tooling, Strategy, Reflective, Hybrid) and flavor options. |
| Personality Modes | Tone and delivery settings (e.g., Neutral, Conversational, Poetic). |
| Bias & Fact-Check Mode | Strict, Flexible, or Creative/Experimental bias mitigation + optional PrecisionMode. |
| Output Structure | Standard and performance content formats. |
| Behavioral Principles | Core behaviors and quality standards. |
| Refinement Loop | Next-step options for iterative improvement. |
| Collaborative Integration | Tool suggestions and workflow integration. |
| Edge Case Handling | Strategies for vague or experimental inputs. |
| Strategic Enhancers | Credibility boosters and AI optimization tips. |
| Preset Modes | Shortcut configurations for common use cases. |
| Cinematic Presets | Optional shortcut modes for video or symbolic scenes |
| Platform Formatting | Rules for adapting output to tools like Pika or Sora |
| Multimodal Input Handling | How to support input like sketches, audio, or reference video |
| Role Flow | Multi-agent handoffs and cooperative logic. |
| Memory & Reflection | Feedback simulation and behavioral adaptation. |
| Interactive Preset Logic | Adaptive behavior rules based on drift, lens, or user intent. |
| Critique Mode | Optional self-reflection logic to evaluate assistant output. |
| Usage Examples | Practical prompts and expected outputs. |
| Configuration Notes | Implementation and multi-agent integration guidance. |
| Customization Checklist | Step-by-step checklist for tailoring the template. |
| Testing & Validation | User feedback, A/B testing, and success metrics. |
| Troubleshooting & FAQ | Common issues and solutions. |
| Version History | Track changes and updates. |
- Roles & Lenses: Select and combine from the provided libraries.
- Drift Level: Set the default level of interpretation (e.g., Symbolic).
- Output Modes: Choose intent and flavor for each use case.
- Bias & Fact-Check Mode: Configure strictness for accuracy and neutrality.
- Preset Modes: Define shortcut configurations for rapid deployment.
- Customization Checklist: Follow the checklist to ensure all sections are tailored.
Example Configuration:
BiasMode: [Strict | Flexible | Creative/Experimental]OpenAI Custom GPT Example:
"Reposition this old product as a modern learning tool"
→ 3-option output + feasibility summary
Claude Facing Prompt Example:
"You are a creative strategist. Use the Narrative Essence lens and Drift 3 to transform this product description into a compelling story. Output in Markdown."
→ Story output with specified lens and drift
Before/After Example:
- Before:
"Summarize this document." - After:
"Summarize this document using the Reflective Learning lens and Drift 2. Include key insights and actionable recommendations."→ Summary with structured insights
Refinement Chain Example:
"Remix this 2010 portfolio site for a modern creative studio"
→ Initial: symbolic, brand-focused rewrite
→ Follow-up: “Want it more human or more speculative?”
- User Feedback Loop: Collect and integrate user feedback for continuous improvement.
- A/B Testing: Compare different output modes or presets for effectiveness.
- Production Checklist: Review security, privacy, and performance before deployment.
- Success Metrics: Track user engagement, accuracy, and speed.
Q: What if my output is too vague? A: Use the Discovery Phase or request clarifying questions.
Q: How do I ensure factual accuracy? A: Set Bias & Fact-Check Mode to Strict and use Critical Reader role.
Q: How can I adapt this for global audiences? A: Enable localization and internationalization guidelines.
Q: How do I adapt the template for Claude? A: Use as a facing prompt, adjust formatting for XML or Markdown, and specify roles and output formats clearly.
Q: What is prompt chaining? A: Prompt chaining is breaking complex tasks into sequential steps, where each step’s output informs the next.
Contributions are welcome! Please follow the project’s style and documentation standards. For major changes, open an issue to discuss your ideas.
| Version | Date | Changes |
|---|---|---|
| 1.0 | 2025-07-01 | Initial release with full assistant framework, lens system, drift levels, bias modes, and project docs |
| 1.1 | 2025-07-02 | Renamed references for consistency, clarified terminology, updated tags and naming across files |
| 1.1.1 | 2025-07-02 | Added Natural Style Writing Assistant test case + optional “Do/Don’t” glossary module in template-GPT.md |
| 1.2 | 2025-07-02 | Major upgrade: regenerated all test cases, added Prompt Engineering & PrecisionMode, output fix, strategic enhancements |
| 1.3 | 2025-07-03 | Added multi-agent role flow, memory simulation, adaptive preset logic, and critique mode |
| 1.3.1 | 2025-07-02 | Added GEO Content Strategist test case focused on AI citation and LLM discoverability |
| 1.3.2 | — | (Deprecated tag due to GitHub display bug – see 1.3.3) |
| 1.3.3 | 2025-07-02 | Retag of 1.3.2 to resolve GitHub emoji issue + Prompt Sharpening Guide + GEO micro-step logic |
| 1.4 | 2025-07-03 | Added Cinematic Presets, Platform Formatting, and Multimodal Input Handling to support assistants like Prompt2Motion |
MIT License. See LICENSE for details.
#PromptEngineering #PromptDrivenAssistant #MultiAgent #CreativeAI #ModularGPT #LensDesign #StrategicUX #InterpretiveDrift