Tools Setup

Before You Begin

For this subject you need at least ONE AI coding tool installed:

• Claude Code (from Subject 1)
• Cursor (from Subject 2)
• Codex CLI (from Subject 3)

The techniques in this subject are tool-agnostic — they work everywhere.

Step 1: Choose Your Primary Testing Tool

Any of these work for prompt practice:

Pick whichever you're most comfortable with. The prompting principles transfer across all tools.

Step 2: Create a Prompts Folder

Set up a place to save and iterate on prompts:

mkdir -p ~/prompts/{templates,experiments,library}

Structure:

• templates/ — reusable prompt skeletons
• experiments/ — A/B tests and iterations
• library/ — proven prompts you use regularly

Step 3: Set Up a Comparison Workflow

The key to improving prompts: compare outputs systematically.

A simple comparison workflow:

1. Write prompt version A
2. Run it 3 times, save outputs
3. Modify to version B
4. Run it 3 times, save outputs
5. Compare: which version produces better, more consistent results?

You can do this manually or use a simple script:

# save in ~/prompts/compare.sh
echo "=== Version A ===" > comparison.md
claude "$1" >> comparison.md
echo -e "\n=== Version B ===" >> comparison.md
claude "$2" >> comparison.md

Step 4: Define "Quality" for Your Prompts

Before improving prompts, decide what "better" means:

• Correctness — does the output do what was asked?
• Completeness — does it handle all cases?
• Consistency — do repeated runs produce similar quality?
• Conciseness — is the output the right length (not bloated)?
• Style — does it match your project conventions?

Not all prompts need to maximise all five. A quick question prioritises conciseness; a feature spec prioritises correctness and completeness.