perf(web): window focus probes instead of resyncing; desktop wake reconnects at once

[tusharbhardwaj-bk]

Problem

The client-sync audit (~/perf-audit-bkt3/30-client-sync.md, F6) found two problems.

1. Every alt-tab resynced everything. Fork commit 4cbe7bf made window focus fire application-active unconditionally. That wakeup:
   • probes the connection;
   • restarts the shell subscription, which on the server means a full shell build;
   • restarts every mounted thread subscription (replay stats plus a replay of up to 1,000 events each);
   • flips those threads to "synchronizing".
2. Coming back from sleep waited 15 s on a dead socket. The desktop window stays visible across a sleep, so only the focus probe noticed the dead socket, after its full CONNECTION_PROBE_TIMEOUT (15 s). The OS resume event only reached telemetry (DesktopTelemetryPublisher.ts:187-190).

Fix

Focus probes; a real return resyncs.

• New wakeup reason application-focus, marked in client-runtime/connection/wakeups.ts.
  • The supervisor probes the session for it (15 s timeout) and it cuts a retry wait short, exactly like application-active.
  • It is excluded from shouldResubscribeAfterWakeup, so streams are not restarted.
  • A failed probe still reconnects without the first backoff rung.
• The web platform (apps/web/src/connection/platform.ts, marked block, helper focusWakeup.expbkt3.ts) tracks when the window was blurred or hidden. A focus after ≥ 60 s away emits application-active (resync as before); a shorter absence emits application-focus.
• A visibilitychange to visible still emits application-active (upstream) and clears the tracker, so the focus that follows only probes rather than resyncing twice.

Desktop resume/unlock reconnects at once.

• New fork layer apps/desktop/src/power/SystemResumeForwarder.expbkt3.ts forwards powerMonitor resume and unlock-screen to every window on desktop:system-resumed. A sliding queue of 1 means a resume and an unlock that fire together produce one message.
• Wiring: marked onSystemResumed in preload.ts, an optional onSystemResumed? field in contracts/ipc.ts, and one line in main.ts.
• The web platform maps it to application-active-reconnect, which replaces the lease immediately. The supervisor already interrupts an in-flight probe for that reason.

Surfaces

• Web (browser tabs): focus behaviour changes as above. Visibility and online wakeups are unchanged.
• Desktop: focus change plus resume/unlock. This reaches users only with the next desktop build: the desktop app loads its bundled web UI, so deploying the server does not change connected desktop clients.
• Mobile: unchanged. It has its own AppState wakeups and never emits application-focus.
• Server: unchanged.

Evidence

• client-runtime/src/connection/supervisor.expbkt3.test.ts:
  • predicate checks;
  • application-focus probes the live session without replacing it (1 session, 0 releases);
  • a failed focus probe reconnects with no TestClock advance.
  • Together with the existing supervisor.test.ts: 46 pass.
• apps/web/src/connection/focusWakeup.expbkt3.test.ts, 4 pass:
  • a quick alt-tab probes;
  • ≥ 60 s away resyncs once;
  • a visibility resync is not doubled;
  • a focus with no recorded absence probes.
• vp run typecheck is clean in packages/contracts, packages/client-runtime, apps/desktop and apps/web. Lint is clean and the fork-marker check passes.

No screenshots: nothing visible changes. The effect is fewer "synchronizing" flips and faster reconnects. I have not exercised it in a real desktop client, because that needs a desktop build.

Model/harness: Claude Opus 5.5 via Claude Code in T3 Code.

🤖 Generated with Claude Code

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

ℹ️ No successful main baseline artifact is available yet. This run establishes the initial measurement.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	—	13.8 KiB	—	15.1 KiB	✅
Codex	Thread snapshot wire	—	7.3 KiB	—	7.8 KiB	✅
Codex	Live turn WebSocket wire	—	6.5 KiB	—	7.8 KiB	✅
Codex	Live turn WebSocket decoded	—	57.0 KiB	—	66.4 KiB	✅
Codex	Live turn messages	—	8	—	21	✅
Claude	Total thread wire	—	13.8 KiB	—	15.1 KiB	✅
Claude	Thread snapshot wire	—	7.3 KiB	—	7.8 KiB	✅
Claude	Live turn WebSocket wire	—	6.5 KiB	—	7.8 KiB	✅
Claude	Live turn WebSocket decoded	—	57.9 KiB	—	66.4 KiB	✅
Claude	Live turn messages	—	9	—	21	✅

Baseline: unavailable · PR result: e4860d8 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 114.9 KiB
• Claude decoded thread snapshot: 115.6 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.