fix(tasks): adaptive visibility-aware polling for cloud task list

[posthog]

Problem

Users who step away from the desktop and create cloud tasks elsewhere (e.g. on mobile while on a walk) return to a stale task list — the new tasks don't appear until a manual reload, undermining the cross-device promise of cloud tasks.

Two compounding issues in useTasks:

• Fixed 30s polling regardless of visibility. refetchInterval: 30_000 keeps hammering the API even when the window is hidden, and never adapts when the user is clearly idle on the task list.
• The on-return refetch was silently swallowed. The global staleTime is 5 minutes and refetchOnWindowFocus: true means "refetch on focus if the query is stale". Returning within 5 minutes left the query "fresh", so no refetch fired — and 5 minutes is exactly the window where this hurts. The previous PR description's claim that refetchOnWindowFocus was working in practice was incorrect: it never loaded the new tasks in practice.

Refs #2335 — alternative implementation that adds adaptive backoff and unit-test coverage.

Changes

apps/code/src/renderer/features/tasks/hooks/useTasks.ts

• New useAdaptiveTaskListPolling(queryKey) helper used by useTasks, useTaskSummaries, and useSlackTasks.
• Polling starts at 30s while the window is focused, exponentially backs off (30s → 60s → 120s → 180s) and caps at a 3-minute ceiling.
• Pauses entirely when the window is blurred or the document is hidden (returns false from refetchInterval, plus refetchIntervalInBackground: false).
• On the blur→focus transition, resets the backoff and explicitly invalidateQueries({ queryKey }) so the list refreshes immediately — independent of TanStack's staleness gate.

apps/code/src/renderer/features/tasks/hooks/useTasks.test.tsx (new)

• Asserts the minimum-interval start, the exponential backoff with ceiling, the pause-on-blur behavior, the focus-return invalidate + backoff reset, and that the initial focused mount does not over-invalidate.
• Drives the real focus pathway by mocking document.hasFocus() and dispatching focus / blur events, exercising the same useRendererWindowFocusStore wiring production uses.

How did you test this?

• pnpm --filter code test useTasks.test — 6 tests pass.
• pnpm --filter code test queryClient — existing focus-manager tests still pass.
• pnpm lint clean on the touched files.
• pnpm --filter code typecheck — the new code introduces no new errors (pre-existing failures in unrelated @posthog/platform/* and @posthog/git/* modules persist on main).
• I could not run the Electron app in this sandbox, so the on-return refetch hasn't been observed in a real window. Worth a smoke test before un-drafting: blur the window, create a cloud task from another client, return to the desktop, confirm the list refreshes within a beat (from the explicit invalidate) rather than waiting up to 3 minutes for the next poll.

Publish to changelog?

no

---

Created with PostHog Code