Show HN: A prompt to make LLMs write property-based tests (github.com)

A new tool named *property-first-testing* has been introduced to enhance the effectiveness of large language models (LLMs) in generating property-based tests. This approach encourages engineers to design tests by breaking down a system into orthogonal properties rather than simply assessing feature-based behaviors. By employing a structured workflow, it allows testing to focus on verifying intrinsic properties, such as invariants, under various conditions, thereby countering the common issue where LLMs primarily generate superficial, example-driven tests that often overlook deeper correctness concerns. The significance of this innovation lies in its potential to improve software testing practices, particularly in complex domains where correctness cannot solely be derived from specific input/output examples. The proposed method champions a dual-tier testing structure—which includes basis tests for individual properties and TLA+-style span tests that account for all composed states—ensuring that both localized behaviors and overall system integrity are thoroughly validated. This paradigm shift not only enhances test coverage but also reinforces the robustness of software systems by ensuring that they comply with underlying correctness principles rather than merely passing conventional tests.