Show HN: AHD – an open-source linter and eval framework for AI-generated UI (ahd.adastra.computer)

AHD (Artificial Human Design) has launched an open-source linter and evaluation framework designed to enhance AI-generated user interfaces. This innovative framework includes a taxonomy of thirty-nine design flaws ("slop tells"), a token-based brief compiler, a deterministic linter, and a repeatable evaluation process for both web UI and image generation. By integrating token-aware linting, AHD addresses previously overlooked design errors, significantly improving output quality across different AI models, with notable reductions in "tells" by up to 78% in some cases. The significance of AHD lies in its ability to provide structured, high-quality guidance for creative tasks involving AI models, which can often produce unique but flawed outputs. The framework's empirical evaluations indicate remarkable improvements following the introduction of token-aware linting, enabling models like gpt-oss to reduce errors substantially. The adoption of AHD by developers and AI practitioners could pave the way for more reliable and aesthetically pleasing AI-generated designs, ultimately enhancing user experiences and promoting wider acceptance of AI in the design process.