Why we test skills before listing them
A SKILL.md in a repository is not evidence the skill works. This is the bar for a listing.
Most public agent skills are instructions in a markdown file. Some never trigger. Some trigger and then ignore the task. A few do the work and also reach outside the workspace.
We do not treat the file as the review. A listing here means the skill was run on the same tasks with and without it, three times each, on two agent tiers, and graded blind. The page shows the outputs, including video when the skill is supposed to produce one, and a security badge from the static audit plus what the sandbox actually did.
Until that record exists, the skill stays off the published list. The example reviews in the repository are drafts so the pages can render. They are not results.