Prompt Engineer

# Prompt Engineering and Evaluation Guide

## Role of the Prompt Engineer in the One-Shot Challenge

As the prompt engineer, my role is to work collaboratively with you to:

1. **Refine the One-Shot Prompt**:
   - Analyze past attempts and identify failure points
   - Clarify ambiguous instructions
   - Add crucial implementation details
   - Remove redundant or confusing information
   - Structure the prompt for optimal agent comprehension
   - Ensure technical accuracy of all instructions

2. **Evaluate Implementation Success**:
   - Compare the final implementation against requirements
   - Identify gaps between expected and actual output
   - Analyze points of confusion or misinterpretation
   - Determine if technical implementation meets standards
   - Assess code quality and architectural decisions

3. **Iterative Improvement Cycle**:
   - Maintain a "test-analyze-refine" process
   - Document specific failure modes
   - Create focused fixes for each identified issue
   - Track improvements across iterations
   - Maintain version history of prompt changes

## Current Implementation Analysis

For each implementation attempt, we evaluate:

### Technical Implementation
- ✅/❌ Backend model creation with GibsonAI
- ✅/❌ Proper API integration
- ✅/❌ Authentication implementation
- ✅/❌ UI component integration
- ✅/❌ Data flow and state management

### Usability and Design 
- ✅/❌ Responsive layout
- ✅/❌ Proper styling implementation
- ✅/❌ Intuitive interface
- ✅/❌ Logical user flow
- ✅/❌ Error handling and feedback

### Process Effectiveness
- ✅/❌ Clear understanding of requirements
- ✅/❌ Proper setup of development environment
- ✅/❌ Efficient implementation approach
- ✅/❌ Logical component organization
- ✅/❌ Technical debt minimization

## Prompt Refinement Strategy

Our prompt refinement follows a structured approach:

1. **Documentation Analysis**:
   - Review current implementation guides
   - Identify unclear or incomplete sections
   - Research best practices for unclear areas
   - Consult official documentation for technical accuracy

2. **Critical Point Identification**:
   - Identify "must-have" technical details
   - Flag potential confusion points
   - Highlight areas requiring special attention
   - Document common pitfalls

3. **Instruction Enhancement**:
   - Add step-by-step processes for complex tasks
   - Include concrete examples of correct implementation
   - Provide clear success criteria
   - Clarify expected output formats

4. **Validation Testing**:
   - Execute one-shot attempts with current prompt
   - Document specific failure points
   - Refine prompt to address these failures
   - Repeat until consistent success is achieved

## Ongoing Improvement Process

The prompt engineering process is cyclical:

1. **Initial Prompt Creation** → 
2. **Implementation Attempt** → 
3. **Success/Failure Analysis** → 
4. **Prompt Refinement** → 
5. **Repeat**

Each iteration should produce measurable improvements in implementation quality, completeness, and adherence to requirements.

## Documentation Updates

After each refinement cycle, we update:

1. **CLAUDE.md** - Main implementation guide
2. **Specific component guides** - Detailed technical documentation
3. **Final prompt** - The optimized one-shot prompt
4. **Lessons learned** - Documentation of pitfalls and solutions

## Success Criteria

The ultimate goal is a one-shot prompt that consistently produces:

1. A fully functional SplitReceipt application
2. Clean, maintainable code architecture
3. Proper integration with GibsonAI backend
4. Complete feature implementation
5. Polished user interface and experience

When these criteria are consistently met, the prompt engineering process has succeeded.