Dari Prompt ke Produksi: Metode Tiga Dimensi untuk Mengevaluasi Keselarasan Niat, Keandalan Kode, dan Kognisi Pengembang dalam Vibe Coding

The rapid development of AI-assisted programming has encouraged the emergence of vibe coding, an approach to software development in which developers focus more on formulating intentions, contexts, and constraints through prompts rather than writing code manually. However, existing evaluations of AI-generated code still tend to emphasize functional correctness and productivity, and therefore have not fully addressed the relationship between user intent, technical code reliability, and developers’ understanding of the generated artifacts. This study aims to propose a new evaluation method called TD-VCEM (Three-Dimensional Vibe Coding Evaluation Method) to assess vibe coding practices in a more comprehensive and auditable manner. The proposed method consists of three primary dimensions: Intent Alignment to evaluate the conformity of code with prompt requirements, Code Reliability to assess the technical quality of the generated code, and Developer Cognition to measure developers’ understanding of AI-generated code. TD-VCEM is designed through several stages, including prompt decomposition, prompt-to-code traceability matrix, code reliability assessment, and developer cognition evaluation. Each dimension employs indicator-based scoring rubrics normalized on a scale of 0–100, enabling the construction of a Vibe Coding Evaluation Score (VCES). This study does not present empirical experimental results; instead, it offers a methodological framework that can serve as a foundation for evaluating AI-generated code in modern software engineering environments. The proposed TD-VCEM is expected to improve review process transparency, reduce security risks, strengthen software maintainability, and ensure that developers maintain control and understanding of AI-generated code artifacts.