VS Code extension app-server crashes with SIGTRAP after creating a local conversation

Summary

The Codex VS Code extension's local app-server exits with SIGTRAP shortly after a new local conversation is created. The extension then reports Codex process is not available and cannot submit turns.

Environment

• macOS
• VS Code
• Codex VS Code extension 26.616.51431 (updated from 26.5616.51431; same behavior)
• Local Codex runtime paired with the desktop app

Steps to reproduce

1. Start VS Code with the Codex extension enabled.
2. Open Codex and create a new local conversation.
3. Wait a few seconds, or attempt to submit a prompt.

Expected behavior

The local conversation remains available and accepts turns.

Actual behavior

The extension log shows the conversation being created, followed by repeated rollout-list warnings and a fatal app-server exit:

Conversation created conversationId=...
codex_rollout::list: state db discrepancy during find_thread_path_by_id_str_in_subdir: falling_back
Codex app-server process exited unexpectedly (code=unknown, signal=SIGTRAP)

After the exit, subsequent requests fail with Codex process is not available.

Troubleshooting already performed

• Restarted VS Code.
• Updated the extension to 26.616.51431.
• Rebuilt the local ~/.codex/state_5.sqlite index.
• Removed all active rollout transcripts with unmatched custom-tool-call outputs from the session scan tree.
• Confirmed the active session tree had zero incomplete transcripts before restarting.

The crash still happens on a freshly created conversation with clean local state, so it appears the app-server treats the fallback condition as fatal.