The transparency rule in practice
"Every listing shows its original prompt" sounds like a small policy until you watch it work. A buyer can read a Skill's actual instructions before spending anything, which changes the psychology of the purchase from gambling to evaluation. A creator, knowing the prompt is visible, invests in genuine engineering — structure, variables, edge-case handling — instead of obscurity. And a competitor's Skill being readable raises the floor for everyone: the fastest way to improve a weak category is that everyone can see what good looks like.
The same rule shapes disputes. When a buyer says a Skill does not do what was promised, the answer is not a judgment call — the prompt is right there, the preview is right there, and the changelog records every claim the listing ever made. Refund decisions under 72 hours are fast precisely because the evidence never needs to be reconstructed.
How curation decisions are made
Submissions are reviewed against a checklist that fits on one screen. Does the Skill produce the output its description promises, in the format promised? Is the original prompt included and readable? Do the previews show real output, not aspirational mockups? Is the category correct? Is the price consistent with what the preview demonstrates? A submission that fails gets the specific reasons and an invitation to resubmit — most rejections are fixed within a day, and the resubmit rate is a metric the curation team tracks the way support tracks response time.
Beyond the checklist, curation watches behavior after launch: install-to-completion rates, first-week retention, and refund patterns. Listings that quietly mislead — a beautiful preview wrapped around a hollow prompt — surface in those numbers, and they come down. The catalog stays small enough that this review is honest; we would rather list four hundred Skills that earn their place than forty thousand that game it.
Skills versus plugins versus one-off prompts
A one-off prompt is a message: brilliant in the moment, unversioned, and unsharable except by copy-paste folklore. A plugin extends an application. A Skill sits in a different place — it is agent-agnostic, text-only, versioned, and sold on its output rather than its integrations. That positioning is why SkillKit does not lock installs to one runtime: the same package pastes into Claude, Cursor or any prompt-capable agent, which keeps buyers portable and keeps the market honest, because a Skill competes on results, not on lock-in.
Community and creators
The creator community is the marketplace's actual product. The 70% share gets the headlines, but the quieter mechanisms matter more day to day: the changelog culture that lets a Skill improve without invalidating its owners, the custom-request pipeline that turns buyer needs into creator opportunities, the invite override that pays the people who grow the community rather than an ad network. We are building the place where prompt engineering stops being an unpaid hobby and starts being a profession — one listing at a time.