What

Re-clicking Login with Nexu account after closing the authorization browser tab now starts a fresh login flow instead of getting stuck for 5 minutes.

Why

Closes #865. When a user started the Nexu cloud-account login and closed the authorization browser tab without completing it, the controller kept polling /api/auth/device-poll for up to 5 minutes and every subsequent click on Login was rejected with "Connection attempt already in progress". The frontend also deliberately short-circuited on that error when status.polling === true, so the UI sat frozen on the waiting spinner and users had a dead-end state with no recovery path.

How

Two small, targeted changes:

Controller (apps/controller/src/store/nexu-config-store.ts)

• New private helper abortDesktopCloudPolling() that aborts pollingState.abortController and clears the reference.
• connectDesktopCloud() no longer errors when a poll is already in flight. Instead it calls the helper, clears persisted desktop cloud state, and falls through to a fresh device-register + new poll. The "Already connected" guard is preserved.
• The three existing inline abort sites (setDesktopCloudProfiles, switchDesktopCloudProfile, disconnectDesktopCloud) now use the shared helper.
• pollDesktopCloudAuthorization() already handles signal.aborted at every await point and returns silently, so aborting an in-flight poll is race-free.

Web (apps/web/src/pages/welcome.tsx)

• Removed the if (data?.error === "Connection attempt already in progress") { ... } dead-end block in handleAccountLogin. Any residual occurrence now falls through to the existing generic data?.error recovery branch that calls disconnect + reconnect.

Deliberately out of scope: the 100-attempt × 3s polling budget, adding an explicit "Cancel" button, any cloud-server-side TTL changes.

Affected areas

• Desktop app (Electron shell)
• Controller (backend / API)
• Web dashboard (React UI)
• Web dashboard (React UI)
• OpenClaw runtime
• Skills
• Shared schemas / packages
• Build / CI / Tooling

Checklist

• pnpm typecheck passes (controller + web)
• pnpm lint passes
• pnpm test passes — pre-existing failures on origin/main in sessions-runtime.test.ts, openclaw-config-compiler.test.ts, openclaw-sync.test.ts, route-compat.test.ts, and one unrelated nexu-config-store.test.ts case (imports cloud profiles and switches active profile while clearing cloud auth). None of the failures touch the cloud-login flow — they reproduce on pristine origin/main without this patch.
• pnpm generate-types run (if API routes/schemas changed) — n/a, no API routes or schemas changed
• No credentials or tokens in code or logs
• No any types introduced

Notes for reviewers

Manual repro to verify the fix:

1. pnpm dev start
2. Welcome → Login with Nexu account → external browser opens the authorization page → desktop UI enters waiting state.
3. Close the browser tab without authorizing.
4. Click Login with Nexu account again.
   • Expected (fixed): a new browser tab opens with a fresh device_id, desktop UI re-enters waiting state immediately, no error toast.
   • Before fix: nothing happens / "Connection attempt already in progress" persists for up to 5 minutes.
5. Complete authorization in the new tab and verify navigation to /workspace.
6. Tail pnpm dev logs controller to confirm two device-register POSTs and that the first poll loop exited via the aborted signal without writing state afterward.

Please also double-check: mid-poll Disconnect still works (now uses the shared abortDesktopCloudPolling helper instead of inline abort).