PR #23025 runtime verification, 2026-10-28, Linux Boat VM
main: 36aea583bb65a04a62dea94074bb0ad137811df1
PR:   3693c958e8002b9f31f598a4cf43854032bc956b

Both checkouts installed with vp i. One dev server at a time, isolated homes.
main server/web ports: 16733 / 8693
PR server/web ports:   13773 / 5733
Browser: repo Playwright 1.60.0, /usr/bin/google-chrome.
SSH: real local OpenSSH server behind transparent TCP proxy, port 22225.
No mock responses, seeded projects, fixtures, or demo provider.

Real client flow:
Settings -> Devices (main) / Integrations -> Devices (PR)
Add host: "ssh keepalive verification", target "pr23025-host".
Test connection: passed; iOS unavailable, Android adb available.
Save host, enable Device hub. App bootstraps actual expo-device-hub over SSH.
Drop traffic after SSH authentication and bootstrap input, preserving the TCP socket.

MAIN
03:15:27.813Z bootstrap SSH connection opened.
03:15:34.099Z proxy began discarding traffic in both directions.
03:17:19Z     ssh PID 18659 still alive (command elapsed 1m51s),
             105 seconds after traffic loss. Client still waiting.
Reloading settings cancelled the request; main server then stopped.
No claim about waiting the full 1,300-second bootstrap budget.

PR
03:26:00.582Z bootstrap SSH connection opened.
03:26:00.932Z proxy dropped traffic after 22,215 client bytes.
SSH argv included -o ServerAliveInterval=15 -o ServerAliveCountMax=3.
03:26:59.942Z server logged ssh.command.failed, exitCode 255:
             "Timeout, server 127.0.0.1 not responding."
59.010 seconds from traffic drop to SSH failure.
Client changed from installing/waiting to Connection failed + Retry.
03:27:38.465Z proxy restored; clicked Retry in the real client.
Remote hub started and forwarded successfully. Device discovery returned
missing-emulator diagnostics instead of a stalled connection.

Real provider path:
Added /home/user/t3code through the client's Local folder project flow.
Selected Codex gpt-6.1-sol and submitted real turns through the composer.
Provider terminal returned the real root and exact PR SHA.
Provider executed runSshCommand against the real SSH server:
  remoteCommandArgs: ["sleep 65; printf healthy-silent-65s"]
  timeoutMs: 900000
  output: {"elapsedMs":65195,"stdout":"healthy-silent-65s","stderr":""}
  exit status: 0 (under timeout 110)
After enabling Agent device access, a NEW Codex session used device_list:
  host: ssh keepalive verification, status: ready
  Android available, iOS unavailable, 0 devices.
  Android discovery incomplete: emulator -list-avds exit 127.
Existing agent sessions retained their disabled device-access setting.
No device was opened or booted.

Checks on task-owned Blacksmith worker:
vp test run packages/ssh/src/command.test.ts packages/ssh/src/tunnel.test.ts apps/server/src/device/SshDeviceHost.test.ts
  3 files passed, 16 tests passed.
vp run check:changed 0a3b561bc0901d8fa47e430fc9d8593918cde978
  format, lint, knip, 4 affected-package typechecks, 8 adjacent tests passed.
The explicit ref is the PR merge base; the worker starts with shallow history.
Hosted PR checks reported 7 successful, 0 failing, 0 pending.

Unverified: native desktop SSH environment recovery and archive cold installs;
macOS/Windows execution; native iOS/Android clients and booted devices;
ControlMaster-shared SSH fault timing, external VPS/network topologies.
This VM lacks emulator/avdmanager and macOS/Xcode.
No source changes or pushes were needed.