[Bug]: Outdated Codex CLI shows "probe failed: exited with code 1" instead of the compatibility warning #20351

Before submitting

• I searched existing issues and did not find a duplicate.
• I included enough detail to reproduce or investigate the problem.

Area

apps/server

Steps to reproduce

1. Have an old Codex CLI first on PATH. Mine is codex 0.1.4161006, the original npm @openai/codex, installed under an old nvm Node prefix.
2. Start T3 Code (vp run dev on main) with the Codex provider enabled (the default).
3. Open Settings → Providers, or a new thread.

Expected behavior

The model manifest already marks Codex <0.149.0 as broken and recommends >=0.159.0. T3 Code should detect the installed version and show that advisory ("known to be incompatible… Use >=0.159.0"), ideally with the Update now action.

This used to work. #404 added a minimum Codex version check that fixed #360. codexCliVersion.ts and that check were removed in the effect-codex-app-server rewrite (#1942).

Actual behavior

The Codex provider shows only:

Codex app-server provider probe failed: Codex App Server process exited with code 1.

Nothing says the CLI is outdated. The old CLI treats app-server as a prompt, prints "Missing OpenAI API key." and exits 1.

The cause seems to be in checkCodexProviderStatus (apps/server/src/provider/CodexProvider.ts). The only source of the Codex version is the userAgent in the app-server initialize response. A CLI that can't complete that handshake gets version: null in all three failure branches (spawn error, non-zero exit, timeout). Both features that would explain the problem are then skipped:

• applyProviderCompatibility (apps/server/src/provider/providerCompatibility.ts) resolves status unknown for a null version, so the manifest's broken range never applies.
• enrichProviderSnapshotWithVersionAdvisory (packages/provider-core/src/server/maintenanceResolver.ts) only looks up the latest version when snapshot.version is set, so no update is offered.

Impact

Minor bug or occasional failure

Version or commit

main @ e58a485

Environment

macOS 26.6.2, Node 24.22.0, dev server (vp run dev), codex 0.1.4161006 resolved from PATH.

Logs or stack traces

$ codex --version
0.1.4161006
$ codex app-server </dev/null; echo "exit: $?"

Missing OpenAI API key.

Set the environment variable OPENAI_API_KEY and re-run this command.
exit: 1

Workaround

Update or remove the old CLI so a current codex is first on PATH, or set Settings → Providers → Codex → Binary path to a current executable. Disabling the Codex provider also hides the error.

Possible fix (happy to send a PR if this direction is OK)

When the app-server probe fails without a spawn error, run codex --version and set its version in the failed snapshot. The fix would be server-only. The client already handles this case: getProviderStatusMessage in apps/web/src/components/chat/ProviderStatusBanner.tsx shows a broken advisory's message ahead of status.message, even when the status is error. With a version set, the existing compatibility and update-advisory code would show "known to be incompatible with this T3 Code release. Use >=0.159.0." and offer the update.

This only affects CLIs that fail the handshake. Versions that answer initialize already get their version and advisory. A broken install whose --version also fails would still show the generic message, which seems fine.