Topic: Learning notes
Standard Prompts: Start with a Clear Request
A short Practical Prompt Engineering note: improve a newsletter-form request with context, constraints, targeted clarification, and concrete checks.
Practical Prompt Engineering on Frontend Masters defines a standard prompt as a direct question or instruction to an AI. “Explain this function” is already a starting point, without arranging a multi-turn workflow or supplying example answers. Course explanation
Following the introduction to prompt engineering, this short note uses an invented form scenario to practice making a request clear.
Turn topic keywords into an actionable request
React form email names a topic, but leaves the task open: explain it, debug something, or build a feature? For a newsletter form, the request could be:
Add a newsletter subscription form to the existing React + TypeScript project.
Include only a required email field, with feedback for invalid formats.
Show a confirmation message after a successful subscription.
Reuse existing form components, validation conventions, and APIs; add no packages.
Read the relevant files before implementing, then report changes and check results.
If you cannot find the subscription API contract, first ask for its path
and success/error response formats.This version adds three kinds of information:
| Information | In this example |
|---|---|
| Task and behavior | Add a subscription form, validate email, show success feedback |
| Relevant context and constraints | Existing stack, components, and API; no new packages |
| What to report afterward | Changes and actual verification results |
The course recommends asking complete questions as you would when seeking help from another person. Paragraphs, lists, and short phrases can all communicate a clear task. Anthropic’s prompting guidance likewise emphasizes explicit output requirements and constraints, with ordered steps where sequence matters.
Clarify consequential gaps before implementation
The final instruction addresses a gap that would affect the implementation. If the repository has no subscription API contract, ask for it rather than inventing an endpoint. If the contract is already clear, reuse it without asking again just to satisfy a question-first routine.
This gives the assistant a clearer rule for handling missing information. Compliance still needs checking. Adding “ask if unclear” does not ensure the model notices every ambiguity or returns identical wording each time.
Check the answer against completion criteria
Evaluate three concrete cases: an empty email is blocked, an invalid format produces feedback, and a valid submission follows the API contract and shows confirmation on success. Reported verification also needs supporting evidence.
The same prompt can still produce different answers. Google’s generation-parameter documentation explains sampling and allows for variation even at temperature 0. The Temperature and Top P note covers this distinction in more detail.
When comparing prompt versions, retain the model, full context, and settings, then evaluate the same cases. That identifies missing requirements more usefully than judging a single response by how fluent it sounds.
References and scope
- Sabrina Goldfarb, course slides, page 8, and the Standard Prompt lesson.
- Anthropic, Prompting best practices: clear instructions, context, and output requirements.
- Google, Content generation parameters: output variation and reproducibility limits.
Sources checked on 2026-10-15. The subscription prompt and checks are invented examples. No model or form tests were run, so this comparison does not establish an improvement in results.