leonxlnx/taste-skill v2.1.1 is an update to a skill repository that helps coding agents use reference images and browser checks when designing screens. According to the GitHub release notes, this version reorganized how skills run in order to reduce cases where the installed agent picks an existing skill by mistake. The seven existing skills, including soft, minimalist, and brutalist, have become opt-in and now run only when they are requested by name. In other words, they are no longer selected by default simply because they are installed.
The screen redesign feature has been absorbed into taste-skill. When working, it keeps the existing stack, content, and the parts that already work, and then fixes things in this order: typography, color, frames, states, and shared components. Showing the changes with before-and-after screenshots was also part of the release. The image generation feature has been split off into a separate repository. The center of this change is therefore twofold: defining the order of screen work and limiting the execution of existing skills to explicit requests. It gathers the features in one place while separating out image generation.
However, the provided release information contains no comparison results for screens that went through this procedure, and no performance measurements. The fact that it asks for screenshots only means a procedure for showing the changes; it does not prove the quality of the results. No information about the license or the primary language was given either. So v2.1.1 can be seen as an update that tidies up the scope of use and the workflow, but whether that procedure improves design outcomes is hard to judge from this information alone.