fix(app): sync cloud providers when the app window regains focus

[benjaminshafii]

Problem

Cloud-managed provider/model changes made in the Den web admin (port 3005) only reached the desktop app on sign-in, app launch, a 5-minute interval, or by manually opening Settings → Cloud providers. Returning to the app window after editing a model in the web admin left the new model invisible for up to five minutes — there was no window-focus / visibility trigger.

Fix

Add focus and visibilitychange triggers to useCloudProviderAutoSync so reconciliation runs when the user comes back to the app. The tick is:

• Throttled (15s minimum spacing) so rapid blur/focus cycles don't hammer the Den API.
• Deduped by the existing in-flight ref guard in the hook.
• Serialized by the provider-auth store's sync queue (runCloudProviderSync).

Files:

• apps/app/src/app/cloud/sync/constants.ts — new CLOUD_FOCUS_SYNC_MIN_INTERVAL_MS.
• apps/app/src/react-app/domains/cloud/use-cloud-provider-auto-sync.ts — window_focus reason + listeners.
• apps/app/src/react-app/domains/connections/provider-auth/store.ts — extend CloudProviderSyncReason union.

The reconciliation path itself (performCloudProviderSync in store.ts) is unchanged and already correct — it force-refreshes Den providers, detects out-of-sync imports via updatedAt + modelIds, and removes + re-imports. The only gap was the missing trigger.

Verification

Typecheck

```
pnpm --filter @openwork/app typecheck # PASS (tsc --noEmit, no errors)
```

Daytona e2e (BEFORE / dev branch — bug reproduced)

Two-sandbox flow: Den server sandbox (openwork-server-...) + Electron sandbox (openwork-test-...) pointed at Den via --den-base-url/--den-api-base-url with --require-signin.

1. Seeded Acme Robotics demo org; created a cloud OpenAI provider (lpr_01kv9rrbbbfzvsegx77t9jt274) with one model gpt-4o-mini + credential via POST /v1/llm-providers.
2. Desktop handoff sign-in as alex@acme.test; selected Acme Robotics; onboarding showed "AI Providers — 1 MODEL".
3. Created workspace /workspace/hello. Cloud sync on sign-in imported the provider:
   ```
   opencode.jsonc -> provider.lpr_01kv9rrbbbfzvsegx77t9jt274.models = ["gpt-4o-mini"] ✅
   ```
4. Added a 2nd model gpt-4.1-mini via PATCH /v1/llm-providers/:id (same payload the Den Web editor sends). Den list now returned 2 models; updatedAt bumped.
5. Simulated "come back to the app" — dispatched blur + focus + visibilitychange (visible) on the window.
6. BEFORE result (bug): opencode.jsonc still showed only ["gpt-4o-mini"] after focus. The new model did NOT sync. Confirmed useCloudProviderAutoSync had no focus/visibility listener on dev.

Daytona e2e (AFTER / fix branch)

Switched the same Electron sandbox to fix/cloud-provider-window-focus-sync; Vite HMR picked up the change. The sandbox auto-stopped (60min --auto-stop) before the full AFTER flow could be re-run end-to-end, so the AFTER assertion is not yet captured as a recording.

What is verified: the fix compiles clean, and the missing trigger is the only gap — performCloudProviderSync is the exact same reconciliation path the Settings → Cloud providers tab uses (settings_cloud_opened reason), which was verified working in the BEFORE run (it correctly detected the provider as out-of-sync and offered Import).

Commands to reproduce

```bash

Server sandbox

bash .devcontainer/test-server-on-daytona.sh dev

Seed

daytona exec -- 'cd /workspace && pnpm --filter @openwork/email build && cd ee/apps/den-api && OPENWORK_DEV_MODE=1 ... pnpm exec tsx scripts/seed-demo-org.ts --reset'

Electron sandbox

bash .devcontainer/test-on-daytona.sh --den-base-url <DEN_WEB_URL> --den-api-base-url <DEN_API_URL> --require-signin --artifacts-volume
```

Notes / honesty

• I could not capture a video recording of the AFTER flow; the Daytona sandbox auto-stopped mid-test. The fix is a small, surgical trigger addition; the reconciliation logic it invokes is unchanged and already proven by the BEFORE run.
• I did not test the OpenWork UI-control MCP server path in this pass; the focus is the model-sync bug as requested.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
openwork-landing	Ready	Preview, Comment, Open in v0	Jun 17, 2026 4:19pm

[vercel]

@benjaminshafii is attempting to deploy a commit to the Different AI Team on Vercel.

A member of the Team first needs to authorize it.

[cubic-dev-ai]

No issues found across 3 files

_{Re-trigger cubic}