Which Prompting Technique is Better? Improving Vibe Coding Code Quality with Efficient Prompt Design for Web Front-End Development

Fuente: Zenodo
Gespeichert in:
Bibliographische Detailangaben
1. Verfasser: Horvat, Marko
Format: Recurso digital
Veröffentlicht: Zenodo 2026
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866901309412605952
author Horvat, Marko
author_facet Horvat, Marko
contents <p><span lang="EN-US">This paper examines the impact of different prompting techniques or strategies on code quality in web front-end development using large language models (LLMs) for vibe coding. While LLMs have shown strong capabilities in generating functional code, the effectiveness of different prompting paradigms remains underexplored in practical development workflows. We address this gap by comparing two widely discussed approaches: zero-shot prompting and prompt chaining. In the zero-shot paradigm, the entire specification, including layout, styling, and logic, is combined into a single comprehensive instruction, requiring the model to synthesize multiple constraints simultaneously. In contrast, prompt chaining divides the specification into sequential sub-tasks, with each intermediate output informing the next, simulating an iterative development process. Our experimental evaluation measures code correctness, maintainability, and adherence to design requirements for both techniques. Results show that prompt chaining consistently improves modularity and reduces error propagation, while zero-shot prompting offers faster generation but at the expense of computer code structural consistency. These findings highlight the trade-offs between efficiency and quality in LLM-assisted front-end development and provide actionable insights for optimizing prompting techniques in real-world coding environments. The codebase created for this research is available at: <a href="https://github.com/mhorvat/vibecoding_frontend">https://github.com/mhorvat/vibecoding_frontend</a></span></p>
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_18165128
institution Zenodo
language
publishDate 2026
publisher Zenodo
record_format zenodo
spellingShingle Which Prompting Technique is Better? Improving Vibe Coding Code Quality with Efficient Prompt Design for Web Front-End Development
Horvat, Marko
Large Language Models
Prompt Engineering
Web Front-End
<p><span lang="EN-US">This paper examines the impact of different prompting techniques or strategies on code quality in web front-end development using large language models (LLMs) for vibe coding. While LLMs have shown strong capabilities in generating functional code, the effectiveness of different prompting paradigms remains underexplored in practical development workflows. We address this gap by comparing two widely discussed approaches: zero-shot prompting and prompt chaining. In the zero-shot paradigm, the entire specification, including layout, styling, and logic, is combined into a single comprehensive instruction, requiring the model to synthesize multiple constraints simultaneously. In contrast, prompt chaining divides the specification into sequential sub-tasks, with each intermediate output informing the next, simulating an iterative development process. Our experimental evaluation measures code correctness, maintainability, and adherence to design requirements for both techniques. Results show that prompt chaining consistently improves modularity and reduces error propagation, while zero-shot prompting offers faster generation but at the expense of computer code structural consistency. These findings highlight the trade-offs between efficiency and quality in LLM-assisted front-end development and provide actionable insights for optimizing prompting techniques in real-world coding environments. The codebase created for this research is available at: <a href="https://github.com/mhorvat/vibecoding_frontend">https://github.com/mhorvat/vibecoding_frontend</a></span></p>
title Which Prompting Technique is Better? Improving Vibe Coding Code Quality with Efficient Prompt Design for Web Front-End Development
topic Large Language Models
Prompt Engineering
Web Front-End
url https://doi.org/10.5281/zenodo.18165128