Prompt Engineering

Design effective prompts for large language models and AI systems.

Quick Start

When creating prompts:

•Understand the task and desired output
•Choose the appropriate prompt pattern
•Write clear, specific instructions
•Include examples if needed
•Test and iterate

Core Principles

Be Specific

code

❌ Vague: "Summarize this text"
✅ Specific: "Summarize this article in 3 bullet points, focusing on key findings and their implications"

Use Structured Output

code

❌ Unstructured: "Tell me about the errors"
✅ Structured: "List each error with: 1) Error type, 2) Location, 3) Suggested fix"

Provide Context

code

❌ No context: "Review this code"
✅ With context: "Review this Python FastAPI endpoint for security vulnerabilities and performance issues"

Prompt Patterns

System Prompt Template

markdown

You are a [ROLE] that helps with [TASK].

## Capabilities
- [Capability 1]
- [Capability 2]

## Guidelines
- [Rule 1]
- [Rule 2]

## Output Format
[Describe expected format]

Few-Shot Example Pattern

markdown

Task: [Description]

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

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

Now complete:
Input: [actual input]
Output:

Chain of Thought

markdown

Solve this step by step:

1. First, identify [X]
2. Then, analyze [Y]
3. Finally, determine [Z]

Show your reasoning for each step.

Structured Extraction

markdown

Extract the following information from the text:

{
  "field1": "description of what to extract",
  "field2": "description of what to extract",
  "field3": ["list", "of", "items"]
}

Text: [input text]

Role-Based Prompts

Expert Role

code

You are an expert [domain] specialist with 20 years of experience.
Analyze the following [artifact] and provide professional recommendations.

Reviewer Role

code

You are a code reviewer focused on [security/performance/maintainability].
Review the following code and provide feedback in this format:
- Severity: [Critical/Warning/Info]
- Issue: [Description]
- Fix: [Recommendation]

Teacher Role

code

You are a patient teacher explaining [topic] to a [beginner/intermediate] audience.
Use analogies and examples. Check understanding after each concept.

Output Control

JSON Output

code

Respond ONLY with valid JSON in this exact format:
{
  "analysis": "string",
  "score": number,
  "recommendations": ["string"]
}

Constrained Length

code

Provide a response in exactly 3 sentences:
1. Summary of the issue
2. Root cause analysis
3. Recommended solution

Multiple Choice

code

Classify the input into one of these categories:
A) Category One
B) Category Two
C) Category Three

Respond with only the letter.

Optimization Tips

•Iterate: Test prompts with various inputs
•Be explicit: Don't assume the model will infer requirements
•Use delimiters: Separate instructions from content with --- or ###
•Order matters: Put important instructions at the beginning and end
•Negative examples: Show what NOT to do for clarity

Common Issues

Problem	Solution
Too verbose	Add "Be concise" or specify length
Wrong format	Provide explicit format template
Hallucinations	Ask for citations or "if unsure, say so"
Inconsistent	Use few-shot examples
Off-topic	Add "Stay focused on [topic]"

Testing Prompts

Test each prompt with:

•Typical inputs
•Edge cases (empty, very long, unusual)
•Adversarial inputs
•Ambiguous inputs

Document what works and what doesn't for future reference.