Prompt Engineering Cheatsheet 📝

Anatomy of a Prompt

graph LR
    A[Complete Prompt] --> B[Goal/Task]
    A --> C[Return Format]
    A --> D[Constraints/Warnings]
    A --> E[Context]
    
    style A fill:#3498db,stroke:#2980b9,color:white,stroke-width:2px
    style B fill:#2ecc71,stroke:#27ae60,color:white,stroke-width:2px
    style C fill:#4a69bd,stroke:#3c6382,color:white,stroke-width:2px
    style D fill:#e74c3c,stroke:#c0392b,color:white,stroke-width:2px
    style E fill:#95a5a6,stroke:#7f8c8d,color:white,stroke-width:2px

Loading

Component	Purpose	Example
🎯 Goal/Task	Define what you want	"List 3 medium-length hikes near SF with unique views"
📋 Return Format	Specify output structure	"For each: name, location, distance, time, features"
⚠️ Warnings	Prevent errors	"Verify trail names and ensure they're currently open"
🌐 Context	Add background info	"I'm an intermediate hiker who prefers less crowded trails"

Example

Source: @benhylak on x

Prompting Strategies

graph TD
    A[Prompting Strategies] --> B[Zero-Shot]
    A --> C[Few-Shot]
    A --> D[Chain of Thought]
    
    style A fill:#3498db,stroke:#2980b9,color:white,stroke-width:2px
    style B fill:#e74c3c,stroke:#c0392b,color:white,stroke-width:2px
    style C fill:#2ecc71,stroke:#27ae60,color:white,stroke-width:2px
    style D fill:#9b59b6,stroke:#8e44ad,color:white,stroke-width:2px

Loading

Zero-Shot vs. Few-Shot

Zero-Shot	Few-Shot
No examples provided	Provides examples first
"Recommend a hiking trail near SF with ocean views"	"Criteria: Family trail, Marin County Answer: Muir Woods... Criteria: Dog-friendly, waterfalls Answer: Alamere Falls... Criteria: Panoramic views, transit access Answer: ?"
Simple but less specific	More verbose but better pattern matching

Chain of Thought (CoT)

Type	Description	Example
Zero-Shot CoT	Add "Let's think step by step"	"What's the best hiking trail for seeing wildflowers in April? Let's think step by step."
Few-Shot CoT	Show reasoning process examples	"Question: Best dog-friendly trail? Reasoning: 1) Check dog policies 2) Find appropriate length... Answer: Waterfall Loop Trail"

Basic vs. Optimized Prompts

graph TD
    subgraph Bad[Bad Prompt]
    BP["Find hiking trails"]
    end
    
    subgraph Good[Good Prompt]
    GP1["GOAL: Recommend 3 moderate trails(4-8 miles)"]
    GP2["near Oakland with views<br>Include: difficulty, location, features"]
    GP3["For context:<br>Im Training for backpacking<br>heres my last runs<br>run 1, from foo to bar. did xyz.<br>run 2, from foo to baz. did xyz."]
    GP1 --> GP2 --> GP3
    end
    
    Bad --> Good
    
    style Bad fill:#990000,stroke:#660000,color:white,stroke-width:2px
    style Good fill:#006600,stroke:#004400,color:white,stroke-width:2px
    style BP fill:#990000,stroke:#660000,color:white
    style GP1 fill:#006600,stroke:#004400,color:white,font-weight:bold
    style GP2 fill:#006600,stroke:#004400,color:white
    style GP3 fill:#006600,stroke:#004400,color:white

Loading

Advanced Techniques: Tree of Thoughts

graph TD
    A[Input] --> B1[Thought 1]
    A --> B2[Thought 2] 
    A --> B3[Thought 3]
    
    B1 --> C1[Thought 1.1]
    B1 --> C2[Thought 1.2]
    B2 --> C3[Thought 2.1]
    B2 --> C4[Thought 2.2]
    B3 --> C5[Thought 3.1]
    
    C1 --> D1[Thought 1.1.1]
    C1 --> D2[Thought 1.1.2]
    C4 --> D3[Thought 2.2.1]
    
    D3 --> Z[Output]
    
    style A fill:#ffffff,stroke:#000000
    style B1 fill:#d9f6d9,stroke:#2e8b57
    style B2 fill:#009900,stroke:#006600,color:white
    style B3 fill:#ffcccb,stroke:#e74c3c
    style C1 fill:#d9f6d9,stroke:#2e8b57
    style C2 fill:#ffcccb,stroke:#e74c3c
    style C3 fill:#ffcccb,stroke:#e74c3c
    style C4 fill:#009900,stroke:#006600,color:white
    style C5 fill:#ffcccb,stroke:#e74c3c
    style D1 fill:#ffcccb,stroke:#e74c3c
    style D2 fill:#ffcccb,stroke:#e74c3c
    style D3 fill:#009900,stroke:#006600,color:white
    style Z fill:#d9f6d9,stroke:#2e8b57

Loading

Prompting Method	Description	Good For
Input-Output	Direct prompting	Simple, straightforward tasks
Chain of Thought (CoT)	Single reasoning path	Step-by-step problems
Tree of Thoughts (ToT)	Multiple reasoning paths	Complex problems requiring exploration

How ToT Works

• Generate multiple thought branches from the input
• Evaluate each branch to determine which is promising (green) vs. dead-end (pink)
• Explore promising paths further while abandoning others
• Continue branching and evaluating until reaching a solution
• Uses search algorithms (breadth-first/depth-first) to explore the tree

Example ToT Prompt

Imagine three different hiking experts are answering this question.
All experts will write down 1 step of their thinking, then share it with the group.
Then all experts will go on to the next step, etc.
If any expert realizes they're wrong at any point, they leave.

The question is: What's the best 3-day backpacking route in Yosemite for someone training for a longer trek?
Consider elevation gain, camping options, water sources, and scenic value.

Optimization Tips

• ✅ Be specific: Clear criteria, measurable targets
• ✅ Structure output: Define format, use bullet points
• ✅ Chunk information: Organize in clear sections
• ✅ Add context: Include relevant preferences/background
• ✅ Use patterns: Test cases, expert perspectives, step-by-step format
• ❌ Avoid: Being vague, contradictory requirements, ignoring model limitations

Remember: Great prompts evolve through iteration!