Prompt Engineering as a Discipline: A Systematic Survey of Patterns, Evaluation, and Design Practice

Prompt engineering — the practice of designing, evaluating, and refining the inputs given to large language models — has grown from informal trial and error into a recognizable discipline with its own patterns, methods, and evaluation practices. Yet the field remains fragmented: practitioner lore, academic studies, and tooling advances are rarely synthesized, and it is often unclear which techniques are robust, which are model-specific folklore, and how prompt quality should be measured at all. This paper presents a systematic survey of prompt engineering as a discipline. It organizes the field into prompt patterns, prompt design methods, and evaluation practices; reviews the evidence for the most widely used techniques, including few-shot prompting, chain-of-thought reasoning, role and context specification, retrieval-augmented prompting, and tool-using prompts; examines the central challenges of brittleness, transferability, security, and evaluation validity; and synthesizes emerging practices such as programmatic prompt pipelines, automatic prompt optimization, and prompt versioning. The survey concludes that prompt engineering is maturing from craft toward engineering discipline, but that this maturation requires shared evaluation standards, documented prompt artifacts, and a clearer distinction between techniques that generalize and techniques that merely fit one model at one moment in time.