A Prompt Engineering Checklist

This checklist distills the strategic principles from this book into a practical tool you can use every time you write, review, or deploy a prompt. Print it, bookmark it, or integrate it into your workflow.

Before Writing the Prompt

Define the goal: What specific outcome does this prompt need to produce?
Identify the audience: Who will see the output? What are their expectations?
Set the token budget: What is the maximum acceptable cost per request?
Choose the model tier: Does this task require a frontier model, or will a smaller model suffice?
Check for existing prompts: Has someone in the organization already solved this problem?

While Writing the Prompt

Be explicit: State every requirement, constraint, and expectation. Assume no shared context.
Structure clearly: Use sections, delimiters, and consistent formatting.
Include examples: Show the model what good output looks like.
Specify the output format: Define exactly how the response should be structured.
Address edge cases: What should the model do when input is ambiguous, incomplete, or adversarial?
Add safety guardrails: Include boundaries for harmful content, injection defense, and graceful failure modes.

After Writing the Prompt

Test with varied inputs: Run at least ten diverse test cases covering happy paths, edge cases, and adversarial inputs.
Test multiple runs: Run the same input at least five times to check consistency.
Evaluate all dimensions: Score correctness, consistency, efficiency, safety, and maintainability.
Review with a colleague: Get a prompt review before deploying to production.
Document the prompt: Record its purpose, version, test results, and known limitations.

In Production

Monitor performance: Track success rate, latency, token usage, and error patterns.
Set up alerts: Define thresholds for anomalies that trigger investigation.
Maintain a changelog: Record every modification with motivation and measured impact.
Regression test on changes: Rerun the full test suite after every prompt modification.
Review periodically: Schedule regular reviews to catch drift and optimization opportunities.

This checklist is a starting point. Adapt it to your organization's needs and revisit it as your prompt engineering practice matures. For hands-on practice applying these principles, explore the Prompt Engineering course on FreeAcademy.