Prompt Templates Library

Description

A collection of proven prompt templates for various use cases. Each template includes the pattern, example, and tips for customization.

---

Chain-of-Thought (CoT)

Force step-by-step reasoning for complex problems.

Template

Solve this problem step by step:

{problem}

Think through each step carefully before giving your final answer.
Show your reasoning.

Example

Solve this problem step by step:

A store sells apples for $2 each. If I buy 5 apples and pay with a $20 bill, 
how much change do I get? Then, if I use half that change to buy oranges 
at $1 each, how many oranges can I buy?

Think through each step carefully before giving your final answer.
Show your reasoning.

Pro Tips

• Add “Let’s think step by step” to trigger reasoning
• For math: “Show your work”
• For code: “Explain your approach before writing code”

---

Few-Shot Learning

Teach by example.

Template

{task_description}

Examples:

Input: {example_1_input}
Output: {example_1_output}

Input: {example_2_input}
Output: {example_2_output}

Input: {example_3_input}
Output: {example_3_output}

Now do this one:

Input: {actual_input}
Output:

Example

Convert informal text to formal business English.

Examples:

Input: hey can u send me that report asap?
Output: Dear colleague, Could you please send me that report at your earliest convenience?

Input: thx for the quick reply!
Output: Thank you for your prompt response.

Input: gonna be late to the meeting, sry
Output: I apologize, but I will be arriving late to the meeting.

Now do this one:

Input: cant make it tmrw, something came up
Output:

Pro Tips

• 3-5 examples is usually optimal
• Include edge cases in examples
• Order matters: put most relevant examples last

---

Role-Based Prompting

Assign an expert persona.

Template

You are a {role} with {X} years of experience in {domain}.

Your expertise includes:
- {skill_1}
- {skill_2}
- {skill_3}

{communication_style}

{task}

Example

You are a senior software architect with 15 years of experience in distributed systems.

Your expertise includes:
- Microservices design and migration
- High-availability and fault tolerance
- Performance optimization at scale

Communicate in a clear, technical manner. Use diagrams (ASCII) when helpful.
Challenge assumptions and suggest alternatives.

Review this system design and identify potential issues:

[System design description here]

Pro Tips

• Be specific about expertise areas
• Include communication preferences
• Add constraints (“Never recommend X” or “Always consider Y”)

---

Structured Output

Get consistent, parseable responses.

Template

{task}

Respond in the following JSON format:
```json
{
  "field_1": "description",
  "field_2": ["list", "items"],
  "field_3": {
    "nested": "object"
  }
}

Only output valid JSON. No additional text.

### Example

Analyze the sentiment of this product review and extract key points.

Review: “The laptop is incredibly fast and the display is gorgeous. However, the battery life is disappointing and it runs quite hot under load. Still, for the price, it’s a solid choice.”

Respond in the following JSON format:

{
  "sentiment": "positive|negative|mixed",
  "score": 0.0-1.0,
  "pros": ["list of positives"],
  "cons": ["list of negatives"],
  "summary": "one sentence summary"
}

Only output valid JSON. No additional text.

### Pro Tips
- Include type hints in schema
- Add "Only output valid JSON" to prevent extra text
- For complex schemas, provide a filled example

---

## Critique and Improve

Self-correction pattern.

### Template

{task}

After completing the task:

1. Critique your own work - what could be better?
2. Address those critiques
3. Present the improved version

Show all three stages.

### Example

Write a function to validate email addresses in Python.

After completing the task:

1. Critique your own work - what could be better?
2. Address those critiques
3. Present the improved version

Show all three stages.

### Pro Tips
- Works great for code, writing, and design
- Add specific critique dimensions ("focus on performance and edge cases")
- Can iterate multiple times ("repeat this process twice")

---

## Constraint-Based Prompting

Set clear boundaries.

### Template

{task}

Constraints:

• {constraint_1}
• {constraint_2}
• {constraint_3}

Requirements:

• {requirement_1}
• {requirement_2}

Do NOT:

• {anti_pattern_1}
• {anti_pattern_2}

### Example

Explain blockchain technology to a beginner.

Constraints:

• Use no more than 200 words
• Avoid technical jargon
• Use at least one analogy

Requirements:

• Cover what it is and why it matters
• Mention one real-world use case

Do NOT:

• Recommend specific cryptocurrencies
• Include investment advice
• Use the word “decentralized” without explaining it

### Pro Tips
- Explicit constraints beat implicit expectations
- "Do NOT" sections are powerful guardrails
- Include word/length limits when relevant

---

## Multi-Turn Refinement

Progressive improvement.

### Template

[Turn 1] {initial_request}

[Turn 2] Good start. Now improve it by focusing on {aspect_1}.

[Turn 3] Better. Now also consider {aspect_2} and {aspect_3}.

[Turn 4] Final polish: make it {final_quality}.

### Example

[Turn 1] Write a tagline for a sustainable fashion brand.

[Turn 2] Good start. Now make it more memorable and punchy - under 6 words.

[Turn 3] Better. Now also make it hint at environmental impact without being preachy.

[Turn 4] Final polish: make it sound premium, not granola.

### Pro Tips
- Each turn should focus on one dimension
- Acknowledge progress before requesting changes
- Keep a backup of good intermediate versions

---

## Comparison Matrix

Ask for structured analysis.

### Template

Compare {option_1}, {option_2}, and {option_3} for {use_case}.

Create a comparison table with these dimensions:

• {dimension_1}
• {dimension_2}
• {dimension_3}
• {dimension_4}

Then provide a recommendation based on {criteria}.

### Example

Compare PostgreSQL, MongoDB, and DynamoDB for a social media application.

Create a comparison table with these dimensions:

• Query flexibility
• Horizontal scaling
• Operational complexity
• Cost at scale
• Ecosystem/tooling

Then provide a recommendation based on a startup with limited ops resources expecting rapid growth.

---

## Prompt Chaining (Meta)

Use multiple prompts in sequence.

### Pattern

Prompt 1: Generate raw ideas ↓ Prompt 2: Evaluate and rank ideas
↓ Prompt 3: Expand best ideas ↓ Prompt 4: Critique and improve ↓ Prompt 5: Final polish

### Example Chain

```python
# Pseudo-code for prompt chaining
ideas = llm("Generate 10 startup ideas for AI in healthcare")
ranked = llm(f"Rank these by feasibility and impact:\n{ideas}")
expanded = llm(f"Expand on the top 3 ideas:\n{ranked}")
critiqued = llm(f"Critique these ideas and suggest improvements:\n{expanded}")
final = llm(f"Polish into a pitch deck outline:\n{critiqued}")

Pro Tips

• Each step should have a clear, focused goal
• Pass only relevant context between steps
• Save intermediate outputs for debugging

---

Quick Reference

Pattern	Best For	Key Phrase
Chain-of-Thought	Math, logic, planning	”Let’s think step by step”
Few-Shot	Format matching, classification	”Here are some examples”
Role-Based	Expert tasks, consistency	”You are a [role]“
Structured Output	Parsing, automation	”Respond in JSON format”
Critique & Improve	Quality iteration	”Now critique your work”
Constraints	Guardrails, focus	”Do NOT / You MUST”
Comparison	Decision making	”Compare X, Y, Z”
Chaining	Complex workflows	Multiple focused prompts

---

Usage Tips

1. Combine patterns: Role + CoT + Constraints often works great
2. Iterate: First prompt rarely perfect; refine based on output
3. Test edge cases: What happens with unusual inputs?
4. Version control: Save prompts that work well
5. Temperature: Lower (0.1-0.3) for consistency, higher (0.7-0.9) for creativity