[Bug]: One failed Codex status check marks the provider unavailable until the next refresh, while sessions still work #25499

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

This is the follow-up that was asked for in #7230 ("timeouts still publish and cache errors ... please open a new issue for that part only").

1. Use Codex with ChatGPT auth and the default 5-minute provider health refresh.
2. Wait for one periodic status check to fail. In my case, the Codex app-server took 15 s and then sent back a JSON-RPC error: workspace routing discovery timed out.
3. Keep using existing Codex threads.

Expected behavior

One failed check does not replace a ready snapshot with an error. T3 checks again soon, and publishes the error only if the failure repeats. An auth failure (unauthenticated) can still show at once.

Actual behavior

The failed check is published and cached as error straight away. The provider shows Unavailable in Settings and in the thread banner. Nothing checks again until the next interval. Running Codex threads keep streaming tokens, but they do not update the status. In practice I go to Settings → Providers and click Refresh, and it recovers in about 3 s.

Timeline from my server trace (checkCodexProviderStatus spans, local time):

The failure came from the first app-server request after initialize, probably account/read. That request has no discovery deadline, so its error fails the whole probe.

Open draft #22794 is in the same area (managedProvider.ts). It may not cover this case, because T3's own timeout did not fire: Codex sent back an error message. A possible fix is in applySnapshotBase: when the previous snapshot was ready and the new probe is not ready (not counting unauthenticated), check once more after a short delay before publishing.

Impact

Codex looks unavailable for up to one refresh interval after one transient upstream failure, while it works. Per #7230, an error snapshot can also block new sessions.

Version or commit

0.1.3-nightly.20261025.3655. The same logic is on main at f0dd86f.

Environment

macOS 26 (Darwin 25.6.0), T3 Code desktop nightly, Codex with ChatGPT auth.

Logs or stack traces

CodexAppServerRequestError: workspace routing discovery timed out
    at CodexAppServerRequestError.fromProtocolError
    at CodexAppServerProtocol.request
    at probeCodexAppServerProvider
    at checkCodexProviderStatus
    at applySnapshot

Workaround

Settings → Providers → Codex → Refresh.