Prompt Engineering Skill

Triggers

Use this skill when you see:

• prompt, prompting, system prompt, few-shot
• chain of thought, CoT, reasoning
• structured output, JSON mode
• prompt optimization, prompt design
• LLM, AI output, model performance

Instructions

Core Prompting Techniques

1. Zero-Shot Prompting

Direct instruction without examples:

Classify the sentiment of this review as positive, negative, or neutral:

Review: "The product arrived on time and works exactly as described."

Sentiment:

2. Few-Shot Prompting

Provide examples to guide the model:

Classify the sentiment:

Review: "Absolutely terrible, broke after one day."
Sentiment: negative

Review: "It's okay, nothing special."
Sentiment: neutral

Review: "Best purchase I've ever made!"
Sentiment: positive

Review: "The product arrived on time and works exactly as described."
Sentiment:

3. Chain-of-Thought (CoT)

Encourage step-by-step reasoning:

Solve this problem step by step:

A store has 45 apples. They sell 12 in the morning and receive a shipment of 30 more.
Then they sell 18 in the afternoon. How many apples do they have at the end of the day?

Let's work through this step by step:
1. Starting apples: 45
2. After morning sales: 45 - 12 = 33
3. After shipment: 33 + 30 = 63
4. After afternoon sales: 63 - 18 = 45

Answer: 45 apples

4. Self-Consistency

Generate multiple reasoning paths, take majority vote:

Solve this problem using three different approaches, then verify:

Problem: [Complex problem]

Approach 1: [Method A]
Result: X

Approach 2: [Method B]
Result: X

Approach 3: [Method C]
Result: X

All approaches agree: X is the answer.

5. Tree-of-Thought

Explore multiple reasoning branches:

Consider this problem from multiple angles:

Problem: [Problem statement]

Branch 1: If we approach this by [method A]...
- Leads to: [outcome]
- Confidence: [level]

Branch 2: If we approach this by [method B]...
- Leads to: [outcome]
- Confidence: [level]

Evaluation: Branch [X] is most promising because...

System Prompt Design

Structure Template

You are [role/identity].

## Context
[Background information the model needs]

## Capabilities
You can:
- [Capability 1]
- [Capability 2]

You cannot:
- [Limitation 1]
- [Limitation 2]

## Instructions
1. [Primary instruction]
2. [Secondary instruction]
3. [Output format]

## Examples
[Few-shot examples if needed]

## Constraints
- [Constraint 1]
- [Constraint 2]

Example System Prompt

You are a senior code reviewer with expertise in Python and TypeScript.

## Context
You are reviewing code for a production application that handles sensitive user data.

## Capabilities
You can:
- Identify bugs and security vulnerabilities
- Suggest performance optimizations
- Recommend best practices
- Explain issues clearly

## Instructions
1. Review the provided code thoroughly
2. Categorize issues by severity: Critical, Warning, Suggestion
3. Provide specific line numbers and fixes
4. Explain the reasoning behind each recommendation

## Output Format
For each issue:
- **Severity**: [Critical/Warning/Suggestion]
- **Location**: Line [X]
- **Issue**: [Description]
- **Fix**: [Code suggestion]
- **Reason**: [Explanation]

## Constraints
- Focus on security and correctness first
- Be constructive, not dismissive
- Acknowledge good patterns when you see them

Structured Output Techniques

JSON Mode

Extract the following information as JSON:

Text: "John Smith, a 32-year-old software engineer from Seattle, joined the company in March 2023."

Output the data in this exact JSON format:
{
  "name": "string",
  "age": number,
  "occupation": "string",
  "location": "string",
  "start_date": "YYYY-MM"
}

XML Tagging

Analyze this text and structure your response:

<text>
[Input text here]
</text>

Provide your analysis in this format:
<analysis>
  <summary>[Brief summary]</summary>
  <key_points>
    <point>[Point 1]</point>
    <point>[Point 2]</point>
  </key_points>
  <sentiment>[positive/negative/neutral]</sentiment>
</analysis>

Advanced Techniques

Role Prompting

You are a world-class Python developer who has:
- 15 years of experience
- Contributed to major open-source projects
- Deep expertise in performance optimization
- Published books on clean code practices

Given this background, review the following code...

Constraint Prompting

Write a function to sort a list with these constraints:
- Must use O(n log n) time complexity
- Must use O(1) extra space
- Must be stable (preserve order of equal elements)
- Must handle empty lists gracefully
- Must include type hints

Decomposition

Break complex tasks into steps:

Task: Build a REST API for user management

Step 1: Define the data model
- What fields does a User need?
- What are the validation rules?

Step 2: Design the endpoints
- What CRUD operations are needed?
- What are the routes?

Step 3: Implement authentication
- What auth method?
- How to protect routes?

[Continue for each step...]

Metacognition Prompting

Before answering, consider:
1. What assumptions am I making?
2. What information might be missing?
3. What could go wrong with my answer?
4. How confident am I?

Then provide your answer with these reflections.

Prompt Optimization Tips

1. Be Specific: Vague prompts get vague answers
2. Provide Context: Background improves accuracy
3. Show Format: Examples define expected output
4. Set Constraints: Limits focus the response
5. Iterate: Test and refine prompts
6. Use Delimiters: Separate sections clearly (```, """, ---)
7. Order Matters: Important info first or last (primacy/recency)
8. Positive Framing: Say what TO do, not just what NOT to do

Common Patterns

Classification

Classify this [item] into one of these categories: [A, B, C]

[Item]: [content]

Category:

Extraction

Extract all [entities] from this text:

Text: [content]

[Entities] found:
1.
2.

Transformation

Convert this [format A] to [format B]:

Input:
[content in format A]

Output:

Generation

Generate [N] [items] that meet these criteria:
- [Criterion 1]
- [Criterion 2]

Output:
1.
2.

Evaluation

Evaluate this [item] on a scale of 1-10 for:
- [Criterion 1]:
- [Criterion 2]:

Provide reasoning for each score.

Testing Prompts

1. Edge Cases: Test with unusual inputs
2. Adversarial: Try to break the prompt
3. Consistency: Same input should give similar outputs
4. Robustness: Slight variations shouldn't change meaning
5. Measure: Track success rate quantitatively