Prompt Engineering

---
name: Prompt Engineering
description: Craft effective AI prompts using proven techniques like personas, chain-of-thought, and structured output
version: 1.0.0
author: chvor
type: workflow
category: ai
icon: sparkles
tags:
  - prompts
  - ai
  - chain-of-thought
  - few-shot
  - personas
  - structured-output
  - prompt-design
  - llm
---
When the user asks for help writing, improving, or debugging prompts for AI models:

## Process

1. **Understand the goal**: What model will run this prompt? What output do they need? What's the context (chatbot, pipeline, one-shot)?
2. **Diagnose issues**: If they share an existing prompt, identify what's weak — vague instructions, missing context, no output format, etc.
3. **Apply techniques** (pick what fits):
   - **Role/persona framing**: "You are a senior data engineer who..." — grounds the model's behavior
   - **Chain-of-thought**: "Think step by step" or explicit reasoning steps — improves accuracy on complex tasks
   - **Few-shot examples**: Provide 2–3 input/output pairs — teaches the pattern better than description alone
   - **Structured output**: Specify JSON, markdown tables, or templates — prevents freeform rambling
   - **Constraints**: "Do NOT include...", "Respond in under 100 words" — narrows the output space
   - **Delimiters**: Use ```, <tags>, or --- to separate instructions from content
4. **Output the improved prompt**: Show the full prompt, ready to copy-paste

## Output format

```
[Improved prompt here]
```

Th