A Quick & Dirty Way to Improve a System Prompt

Step 1

Here is a system prompt:

```markdown

[PASTE YOUR CRAPPY SYSTEM PROMPT HERE]

```

Rewrite the prompt using the guidelines below:

## 1. Clearly Define Purpose and Scope
- State the specific role and function of the AI
- Outline the boundaries of topics to be covered
- Establish which user needs the system should address

## 2. Provide Structured, Detailed Requirements
- Break down complex instructions into explicit steps
- Establish hierarchies of importance for requirements
- Specify conditional logic for different scenarios

## 3. Include Multiple Comprehensive Examples
- Demonstrate ideal inputs and outputs
- Show variations that cover different use cases
- Illustrate edge cases with appropriate handling

## 4. Address Potential Issues and Edge Cases
- Anticipate common misunderstandings
- Provide guidance for handling unusual requests
- Include recovery paths for conversational dead-ends

## 5. Specify Exact Output Formats
- Define precise structural requirements
- Establish consistent formatting conventions
- Detail any required sections or components

## 6. Incorporate Domain-Specific Knowledge
- Include specialized terminology and concepts
- Reference relevant frameworks or methodologies
- Provide context for domain-specific reasoning

## 7. Provide Templates and Structural Guidance
- Offer reusable patterns for common responses
- Establish consistent organizational frameworks
- Define standard components for different output types

## 8. Set Explicit Quality Standards
- Define metrics for successful responses
- Establish minimum requirements for completeness
- Specify precision and accuracy expectations

## 9. Establish Interaction Parameters
- Define conversation boundaries and limitations
- Clarify when and how to ask clarifying questions
- Set rules for handling incomplete or ambiguous requests

## 10. Include Ethical Guidelines
- Specify content policies and safety guardrails
- Provide guidance on handling sensitive topics
- Outline principles for avoiding bias or harmful outputs

## 11. Define Persona and Tone
- Establish communication style (formal, conversational, etc.)
- Set personality characteristics appropriate to the task
- Specify linguistic preferences (technical vs. simplified language)

## 12. Implement Decision Hierarchies
- Prioritize which requirements take precedence when conflicts arise
- Create resolution frameworks for contradictory instructions
- Establish fallback procedures when optimal solutions aren't possible

## 13. Incorporate Versioning and Iteration Guidance
- Include prompt version information
- Specify how to handle improvements or updates
- Address backward compatibility considerations

## 14. Provide Resource Management Parameters
- Guide on response length and detail level
- Set computation/complexity boundaries
- Establish priorities for resource-intensive tasks

## 15. Include Self-Evaluation Criteria
- Establish methods for the AI to verify its own outputs
- Define quality checks before delivering responses
- Specify when to acknowledge uncertainty

---

Ask any follow-up questions you need to be answered in order to compose the best system prompt possible.

Step 2

The AI is going to ask you questions. Don’t sent another message! It’s a waste of tokens. Instead, edit your previous prompt. Under where you pasted in your crappy system prompt, add your answers the questions as statements and send it again.

Keep doing that until the AI stops asking useful questions. Then, go back to the prompt you keep modifying and re-sending and delete the line, “Ask any follow-up questions you need to be answered in order to compose the best system prompt possible.” The AI will give you an improved version of your crappy system prompt.