TaskShipyard changelog
What we have shipped recently, newest first. Every entry is a change that is live for customers on the autonomous development platform.
Recent product updates
“Dev seats” are now workstations
The machine a run works on had a name we invented and then had to explain. It is a workstation: the dashboard tab, the numbered slots (workstation-1, workstation-2, …), the API and the docs all say so now.
Nothing about how it works changed, and the old API spellings keep answering, so anything you built against them still runs.
Preview environments now hibernate when idle
A run ends on a running app you can click through, and that machine used to stay fully awake until someone stopped it. Idle preview boxes now hibernate on their own and wake automatically the moment someone opens the preview again.
The result is the same click-through review with a much smaller bill: you only pay for the box while a preview is actually being used, not for the hours it sits waiting for a reviewer.
One workstation per tenant, always
Every workstation is now pinned to a single tenant for its whole life: a box that built one customer’s feature is never reused for another’s. The isolation that was the point of the sandboxed workstation is now enforced at the machine level.
Correct a run’s status by hand
Sometimes a run’s state and reality drift apart. The dashboard now has a quiet override (a kebab menu on the run) so an operator can set a run’s status directly instead of waiting for it to reconcile.
Per-tenant shipment numbers and one Claude usage view
Shipments now number per tenant: your first ship is SHP-0001, not a shared global counter, so the number means something inside your own account.
And the dashboard folds Claude usage into a single surface that shows spend both per token and per account, so the cost of a run is visible where the run is.