Agent and Skill pages show their own content again

Opening /Agent or /Skill showed a bare page: the right name was missing, the description was missing, and a link to either page shared as an empty card on Slack, Teams or LinkedIn. The published packages behind those pages were fine the whole time — nothing was reading them.

Five top-level pages share a name with something the platform defines internally: Agent, Skill, Harness, Model and Provider. A deployment says whether it stores that content in its database, and when it does, the internal definition steps aside and the stored content owns the page. A recent internal change stopped that setting reaching the part of the platform that reads it, so every deployment behaved as if it had no setting at all: the internal definition took the page and the stored content became unreachable — not just outranked, but invisible to the page, to search, and to anything trying to update it.

The setting now reaches it on every deployment, and a page can no longer be quietly taken over this way. Two related effects are fixed with it: content for those five areas is imported and kept up to date on start-up again, and the model Provider area — which had never been created on affected deployments — now appears.

If you maintain a deployment: nothing to change. Your existing configuration was always correct; it simply was not being read.

Reconnecting…
The connection to the server was interrupted. Trying to restore it…
Trying again…
The connection could not be restored. Reloading the page…
The server was updated. Reloading the page to pick up the latest version.