How we test

1. One standard brief

Each test starts from the same brief: a landing page for Tally, a fictional budgeting app, with a hero, at least three features, a three-tier pricing table and a footer, as a single HTML file. No real metrics, testimonials or logos are provided. Read the full brief.

2. Before: no skill

We generate the page from the brief with no design guidance. That one baseline is the "before" for every listing.

3. After: the skill's published instructions

We take the instructions the skill publishes (its SKILL.md, prompts and rules on GitHub) and apply them ourselves to the same brief. We do not run the skill inside its own agent, plugin or tooling. Scripts, MCP calls, asset pipelines and multi-step workflows the skill may rely on are not executed, so a native install can produce different, sometimes better, results.

4. Screenshot and score

Both pages are screenshotted full-page at 1440px. We score the "after" from 1 to 5 on hierarchy, typography, layout, how well it follows the brief, and whether it avoids generic AI-template patterns. The verdict is one editor's judgement, not a benchmark.

What we don't test

MCP servers, Figma integrations and collections usually can't be judged on a single HTML brief. Those listings say so, and carry no score.

Independence

Featured placement never changes a score or verdict. Spot something wrong? Tell us and we'll retest.