worker_restarted
{ "error": { "code": "worker_restarted", "message": "worker restarted; retry safely" } }HTTP status: 503
Header: Retry-After: 1
A Kifas worker pod was rotated (deploy, scale-down, or health-check failure) while your call was in flight. The platform writes no partial state on restart — your call simply didn’t land.
This is also returned when the sandbox-runner’s per-process semaphore is full and the call would have been queued past the saturation cap.
Retry once after Retry-After seconds. The @kifas/mcp CLI does this automatically.
If you see sustained worker_restarted with no normal calls succeeding between, ping
support — there may be an outage.
Idempotency: every read tool (session.snapshot, session.screenshot, tasks.poll,
run.list, etc.) is safe to retry blindly. Mutating tools (session.click,
session.type) require care — if the call did land before the worker died, the retry
will perform the action twice. Use session.snapshot to confirm state before retrying.