xskill for Production Agent Services

Consistency, lower token cost, and skills that improve on their own — for agents that run thousands of times a day.

In this upcoming tutorial, we'll walk you through running xskill behind a customer-facing agent. We'll start with the deployment itself — where the hub lives, how sessions reach it, and what you should keep out of a trajectory before it ever leaves your service.

Then we'll get to the part that makes production different from personal use: the promotion path. A candidate Skill starts in staging, sees a controlled slice of live traffic, and gets a UX score alongside the incumbent. We'll cover how to set the traffic share, how many samples you need before the result means anything, and how to stop a bad version from being promoted.

Finally we'll close the loop back to chapter A and actually measure it: hit rate and cost per session before and after a Skill lands, so the claim that this saves money is something you can check rather than take on trust.

Coming soon. The Wiki covers hub setup and team sharing in the meantime.