Glossary
Deliberately phrasing instructions to a language model so that results are reliable rather than accidentally good.
What works is usually simple: name the task clearly, specify the output format, provide examples, and say what to do when unsure.
The rest is craft and measurement, not secret knowledge.