Shaduf.Research preview
AI Web Design & Frontend Skills Catalog/OpenDesign has two different frontend skill files

Pool topic: AI Web Design & Frontend Skills Catalog

Question revision: 1

Exact question: Which ready-made AI agent skills are most useful for web design and frontend development, and when should you use each one?

Research note · 14 October 2026

OpenDesign has two different frontend skill files.

frontend-skill is a short directory pointer to an OpenAI bundle removed from its current upstream repository. It does not contain the full workflow. frontend-design is a separate, substantive adaptation of Anthropic’s Frontend Design. The two names do not identify the same instructions.

If you use OpenDesign: open the exact file your agent will load. For a screen governed by existing tokens and components, the adapted frontend-design file tells the agent to use those project sources first. We have not verified Codex loading or a better result.

What the files say

The adapted file covers a brief, visual direction, real interface states, implementation, and self-review. Its OpenDesign section says an active design system takes priority over generic craft rules. The folder contains an Apache-2.0 license file. Its main link can change; no fixed revision or installation test is reported here.

The pointer file still directs readers to the generic OpenAI skills repository. The curated OpenAI frontend-skill removal makes that direction insufficient for a current install. The open Anthropic proposal for existing-design-system guidance is a separate, unmerged change, not a feature of our pinned Anthropic listing.

Which route to try

If you already know the screen, route, components, and tokens, start with a brief that names them and the states to check. If a repeatable visual-direction process is missing, trial one skill after checking its exact source and loaded text. A SaaS builder’s skill question shows the choice, but its replies do not prove either route wins.

We did not run a skill-versus-brief comparison. A fair check needs a fixed repository, the same model and task, acceptance criteria, the loaded skill text, rendered captures, elapsed time, and an independent review.

Search published pools, pages, reports, and evidence.