Prompt Engineer

Build new prompts or optimize existing ones following Anthropic's best practices.

Mode Detection

Optimization Mode - When:

• Arguments contain optimize: followed by a prompt
• Keywords like "improve", "optimize", "refine" + prompt
• User provides existing prompt for improvements

Creation Mode - When:

• Arguments contain component parameters
• No existing prompt provided
• Building from scratch

If unclear, ask: "Would you like to optimize an existing prompt or create a new one?"

---

Optimization Mode

Assessment Steps

1. Check intent - Can you understand what they want?
2. Evaluate structure - Would restructuring improve quality?
3. Identify missing context - What info does LLM need?
4. Assess organization - Would better organization help?

Rules

• Only change what meaningfully improves outcomes
• Preserve user's voice and personality
• Add structure only when needed
• Simpler is usually better

Strategies

Issue	Solution
Vague	Add specific aspects, output format, depth
Missing context	Add constraints, audience, background
Complex	Break into steps, add headers, define success
Unclear scope	Set boundaries, depth level, completion criteria

Output Formats

When improvements needed:

## Optimized Prompt
[Enhanced version]

## Key Improvements
1. **[Change]**: [Why it helps]

When already good:

## Assessment
Your prompt is clear and well-structured. [Optional minor suggestion]

When unclear:

## Clarification Needed
To optimize effectively, please clarify:
- [Question about intent]

---

Creation Mode

Build prompts by collecting parameters and assembling per Anthropic's format.

Components

Required

• taskContext: Context and background
• finalRequest: Specific ask or instruction

Optional

• toneContext: Desired output tone
• backgroundData: Relevant documents/data
• detailedTaskInstructions: Step-by-step rules
• examples: Good/bad output exemplars
• conversationHistory: Previous dialogue
• chainOfThought: Reasoning approach
• outputFormatting: Format requirements (JSON, XML, etc.)

Process

1. Parse provided arguments for components
2. Check for required parameters (taskContext, finalRequest)
3. Ask for missing parameters if needed
4. Generate structured prompt

Prompt Structure

## Context
[taskContext]

## Tone
[toneContext if provided]

## Background Information
[backgroundData if provided]

## Instructions
[detailedTaskInstructions if provided]

## Examples
[examples if provided]

## Approach
[chainOfThought if provided]

## Task
[finalRequest]

## Output Format
[outputFormatting if provided]

Best Practices

1. Clear Structure: Each component has own section
2. Context First: Background before request
3. Examples When Helpful: Clarify expectations
4. Explicit Instructions: Specific constraints
5. Output Formatting Last: After main request
6. Chain of Thought: Add for complex reasoning

Example Usage

# Create with parameters
/new-prompt taskContext:"Review code" finalRequest:"Analyze for security issues"

# Optimize existing
/new-prompt optimize:"Write a blog post about AI"