VS Code extension intermittently reports “Codex process is not available” with gpt-5.6-sol and personality fallback warning

Environment

• IDE: Visual Studio Code on Windows
• Codex extension version: 26.707.31428
• Model: gpt-5.6-sol
• Reasoning effort: ultra
• config.toml contains:

model = "gpt-5.6-sol"
model_reasoning_effort = "ultra"
personality = "none"
service_tier = "default"

[windows]
sandbox = "elevated"

• The Codex UI personality selector only exposes Friendly or Pragmatic. Pragmatic was selected in the UI even though config.toml says personality = "none".

Issue

The VS Code Codex panel intermittently crashes and displays:

Codex process errored: Codex process is not available

codex_protocol::openai_models: Model personality requested but model_messages is missing, falling back to base instructions.
model=gpt-5.6-sol personality=pragmatic

Reloading Codex restores the UI. In at least one occurrence, after reloading and reopening the panel, Codex continued working on the previously assigned goal. This suggests the underlying task/session may still be alive or recoverable while the local extension process/UI connection fails.

Reproduction observed

1. Configure gpt-5.6-sol with model_reasoning_effort = "ultra".
2. Keep personality = "none" in config.toml.
3. Select Pragmatic in the Codex UI because the UI does not provide a None option.
4. Use Codex normally on a codebase.
5. The panel may later show Codex process is not available together with the missing model_messages personality fallback warning.
6. Reload Codex.
7. The task may resume, indicating a recoverable extension/process failure rather than a full task failure.

Expected behavior

• The VS Code extension should not become unavailable because personality metadata is missing.
• If a model does not support personalities, the UI should provide a None option or clearly disable the selector.
• A personality fallback warning should remain non-fatal.
• If the background task remains active, the panel should reconnect without presenting a full process-crash screen.

Additional context

OpenAI status showed only a FedRAMP-specific incident at the time, so this did not appear to be part of a general service outage.

There are existing issues containing the same model_messages is missing warning for other models, but I did not find an exact report combining gpt-5.6-sol, the VS Code extension, the UI/TOML personality mismatch, and the recoverable Codex process is not available state.