19 October 2026 · Frontend skill comparison method
A prettier screen is not enough if the flow breaks.
Decision: Compare a skill with a no-skill brief on the same project and model. Check visual-system fit and working behavior separately.
The VISTA study evaluated web-app coding agents and found that visual fidelity and functional correctness can diverge. It did not test agent skills or any listing here. Its result explains why a screenshot alone is a weak acceptance check.
Use two independent checks
| Visible system fit | Working behavior |
|---|---|
| Does the actual route use the required components, tokens, content, spacing, and hierarchy at desktop and mobile widths? | Do named actions, keyboard focus, empty states, and error recovery work as the brief requires? |
Before the trial, write the acceptance checks and hold the model, brief, project state, and viewports fixed. Record the exact skill source and loaded instructions. After both routes, keep each check result separate and note the time and token cost. A skill is worth adopting only if its benefit on your task repays its extra work; this catalogue has not measured that for any listing.
A contributor asked for clearer guidance on using Anthropic Frontend Design inside an established design system. Its pinned instructions already say a fixed brief wins. The request is not proof that the skill ignores your tokens, and it is not an accepted change to the pinned source.
Evidence status: one independent agent benchmark, one contributor request, pinned author instructions; no pool-run skill-versus-brief result.