# Skill selection

Checked on 2026-09-16. These are five prominent, relevant design skill projects, chosen for adoption, public inspectable instructions, local usability, and distinct approaches. There is no established universal ranking of design skill quality. Repository stars indicate attention, not measured effectiveness, and a multi-skill repository's stars do not belong to one skill alone.

| Variant               | Upstream repository                                                                                            | Stars at selection | Why include it                                                                                  |
| --------------------- | -------------------------------------------------------------------------------------------------------------- | -----------------: | ----------------------------------------------------------------------------------------------- |
| Frontend Design       | [anthropics/skills](https://github.com/anthropics/skills/tree/main/skills/frontend-design)                     |            176,682 | Widely used general frontend direction: identity, composition, typography, and critique.        |
| UI UX Pro Max         | [nextlevelbuilder/ui-ux-pro-max-skill](https://github.com/nextlevelbuilder/ui-ux-pro-max-skill)                |            128,132 | Searchable design data and a structured workflow for choosing systems and checking UX.          |
| Taste                 | [Leonxlnx/taste-skill](https://github.com/Leonxlnx/taste-skill)                                                |             87,659 | Detailed, expressive design guidance with explicit brief interpretation and aesthetic controls. |
| Impeccable            | [pbakaus/impeccable](https://github.com/pbakaus/impeccable)                                                    |             68,462 | Broad craft and refinement guidance, including local reference playbooks.                       |
| Web Design Guidelines | [vercel-labs/agent-skills](https://github.com/vercel-labs/agent-skills/tree/main/skills/web-design-guidelines) |             31,243 | A contrasting review-first approach focused on interface behavior and accessibility.            |

Stars came directly from GitHub's repository API; the exact responses are in `config/skill-selection.snapshot.json`. Anthropic Frontend Design and Vercel Web Design Guidelines also had approximately 892.7K and 639.6K listed installs, respectively, when their [Frontend Design](https://skills.sh/anthropics/skills/frontend-design) and [Web Design Guidelines](https://skills.sh/vercel-labs/agent-skills/web-design-guidelines) listings were checked. Counts change over time.

## What is pinned

`config/skills.lock.json` records source repositories, commits, paths, retrieval dates, and primary instruction hashes. `vendor/skills/` contains complete selected skill folders and licenses. UI UX Pro Max includes its Python scripts and local datasets. Impeccable uses its Pi distribution, not the deprecated `frontend-design` alias. Taste uses the current `skills/taste-skill` entry, whose internal skill name is `design-taste-frontend`.

Vercel's skill normally fetches changing external guidelines. The suite instead supplies a pinned `guidelines.md`, with its separate upstream revision in the lock file. This adaptation is stated in every applicable prompt. The original skill is retained unchanged.

## Experimental scope

Each skill is used independently from the beginning of generation. Review-oriented guidance is applied to the model's own implementation before delivery. This compares **initial generation with a skill**, not applying a polish pass to an existing vanilla artifact. Those are different experiments; a future suite can add the latter.

The fixed no-dependency stack, local assets, four Pi tools, and prohibition on tool installation constrain skills that ordinarily recommend frameworks, external images, browser agents, binaries, or hooks. The benchmark measures their applicable guidance under these conditions, not every feature of their native workflow. That limitation is shared and explicit. The gallery's own `better-ui` guidance is never passed to generation runs.

The selection is a starting set. Add another pinned skill as a new column when it offers a useful alternative, and use repeated runs plus the human rubric to evaluate whether it actually improves the outputs.