There isn't one agent-skill marketplace. There are several general directories aggregating listings from independent sellers, plus a smaller number of storefronts that build and maintain their own catalogue instead of hosting other people's. That split matters more than it looks, because it changes who's actually accountable for a listing being current β and "current" is exactly what you can't verify by looking at a product page.
Why this is worth a checklist
A skill file or MCP tool listing usually references a third-party API: rate limits, free-tier terms, auth flow, sometimes a specific pricing tier. All of that drifts. A listing that was accurate the day it was published can be quietly wrong six months later, and nothing about an aggregator's structure obligates anyone to catch it β the seller who wrote it is the only one with a reason to update it, and update discipline varies seller to seller.
So before you install or pay for one, run it through four checks:
1. Read the raw file before you buy, not after
A skill is plain Markdown (or, for MCP, a tool schema plus a short description). Neither format requires a black box. If a seller won't let you see the actual skill file or MCP tool definition before purchase, that's the tell β there's no technical reason to hide it, since the value is in the tuning and testing, not in obfuscating the prompt. A legitimate listing should be readable up front.
2. Check the update history, not the star rating
A star rating tells you people liked it when they bought it. It says nothing about whether the underlying API it wraps still behaves the way the listing describes. Look for a changelog, a "last verified" date, or commit history if the skill lives in a public repo. No visible update trail is itself a data point.
3. Separate the seller's claim from the third-party's current terms
If a listing claims "free tier covers X requests/month," that's a claim about someone else's product, made at some point in the past. Cross-check it against the third-party's actual current docs before you build a workflow that assumes it's still true. This is the single most common way an otherwise-good skill goes stale: the integration logic is fine, the pricing assumption baked into it isn't.
4. Know whether you're buying from one seller or one of many
On an aggregator, quality is uneven by design β you're evaluating one listing among thousands, each maintained independently. On a single-storefront catalogue, one team is accountable for every kit, which doesn't make it automatically better, but it does mean there's one throat to choke when something breaks instead of trying to track down whichever individual author listed it. Know which one you're looking at; the size of the catalogue tells you nothing about which structure you're in.
The takeaway
None of this is specific to any one platform. It's the same due diligence you'd apply to any dependency you're about to hand credentials to: read the source, check when it was last touched, and verify third-party claims against the third party β not the listing.
If you want to see what "readable before you buy" looks like in practice, AgentKit Works publishes a full sample skill file and gives away its starter kit free, so you can check the bar before spending anything: https://agentkitworks.com/sample-skill
