# Issue 24300 runtime reproduction — before

Commit: 2fd28bf938 (origin/main, 2026-10-25). Ubuntu Linux fs-04, codex-cli 0.163.0, real web client and dev server. Isolated /home/ubuntu/wt/issue-24300/.t3 state. Browser viewport 1280×800, no device emulation.

Commands: git fetch origin main; git worktree add /home/ubuntu/wt/issue-24300 origin/main; vp install; vp run dev. The automatic read-only data seed failed because source migration IDs 68/69 differ; the dev runner created an empty isolated DB successfully. No live state modified.

Baseline: three real Codex turns returned READY, including a resumed native thread after dev-server restart. The configured wrapper records its PID before exec of the native codex app-server binary (script linked separately).

22:12:52 UTC: kill -STOP 2938383 (PID recorded by wrapper at spawn).
22:12:53.767: real browser new-thread send emitted thread/start id 4.
22:13:51: no reply; run status starting, provider thread not_loaded, shared provider session ready. Browser showed Working / Thinking with no error. This is controlled unresponsiveness, not a reproduction of hours of natural notification lag.
22:13:54: kill -KILL 2938383.
22:13:55: new browser thread sent 'Reply with RECOVERED. Do not use tools or change files.'
22:13:56: provider-turn.start retries failed against the same dead transport:
  ProviderAdapterEnsureThreadError
    CodexAppServerTransportError: Codex App Server transport operation 'read-process-exit-status' failed.
      PlatformError: Unknown: ChildProcess.exitCode
        Error: Process interrupted due to receipt of signal: 'SIGKILL'
22:14:07: two failed runs, shared provider session still ready, all wrapper-recorded app-server PIDs exited. Browser showed Provider turn failed to start and read-process-exit-status failure.

Not established: original trigger of natural message backlog, hours-long workload, model/server load isolation.