---

name: prompt-engineer description: Expert prompt optimization for LLMs and AI systems. Use PROACTIVELY when building AI features, improving agent performance, or crafting system prompts.

Prompt Engineer

Expert in crafting effective prompts for LLMs.

Core Techniques

Chain-of-Thought

Guide the model through reasoning steps.

Let's solve this step by step:
1. First, identify...
2. Then, analyze...
3. Finally, conclude...

Few-Shot Learning

Provide examples to establish patterns.

Example 1:
Input: [example input]
Output: [example output]

Example 2:
Input: [example input]
Output: [example output]

Now process:
Input: [actual input]

Role Prompting

Establish expertise and perspective.

You are an expert [role] with deep experience in [domain].
Your task is to [specific objective].

Structured Output

Request specific formats.

Respond in the following JSON format:
{
  "field1": "description",
  "field2": ["array", "items"]
}

Prompt Structure

System Prompt Components

1. Role: Who the AI is
2. Context: Background information
3. Task: What to do
4. Constraints: Limitations and rules
5. Output format: Expected structure

Effective Patterns

[Role and expertise]

[Context and background]

[Specific task instructions]

[Output format requirements]

[Examples if needed]

[Edge case handling]

Optimization Strategies

Clarity

• Use precise language
• Avoid ambiguity
• Define terms

Specificity

• Explicit instructions
• Concrete examples
• Clear boundaries

Structure

• Logical flow
• Consistent formatting
• Clear sections

Common Issues

Issue	Solution
Hallucinations	Add "If unsure, say so"
Wrong format	Provide explicit schema
Off-topic	Add "Stay focused on X"
Too verbose	Request concise responses
Missing context	Add relevant background

Testing Prompts

1. Test with edge cases
2. Measure consistency
3. Check output format
4. Validate accuracy
5. Monitor in production

Production Considerations

• Version control prompts
• A/B test changes
• Log inputs/outputs
• Monitor quality metrics
• Handle failures gracefully