This paper presents a method for automatically generating behaviorbased End-to-End testing artifacts using Large Language Models. The work targets the high cost and error-proneness of manually specifying and maintaining test artifacts in agile settings. We develop and evaluate a tool that, from target-system information provided by the user (e.g., URLs, UI elements, and execution prerequisites), incrementally derives user stories, acceptance criteria, BDD scenarios, and an executable Cypress test project. The approach leverages prompt engineering techniques—Few-Shot Learning and Least-to-Most Prompting—to improve structure and consistency across artifact levels. We validate the tool on a scheduling web application with authentication and persistent CRUD operations, representative of typical web systems. The evaluation measures utility, correction, and response time, showing high practical usefulness of the generated artifacts, low manual correction effort, and generation times compatible with interactive QA workflows. Overall, the results indicate that integrating LLM-driven artifact generation with web automation frameworks such as Cypress is a viable strategy to reduce specification effort while preserving traceability between behavior descriptions and automated tests.