Codex Desktop keeps auto-retrying an auth-failed stale SSH connection instead of backing off cleanly

## Summary

On my Windows machine, Codex Desktop keeps automatically retrying a saved remote SSH connection after authentication has already failed.

This may have been triggered by a stale local remote configuration, but the app appears to keep retrying the failed connection in the background instead of backing off cleanly or requiring explicit user action.

Environment

• Platform: Windows
• Date observed: July 4, 2026

Connection involved

• hostId: remote-ssh-codex-managed:<saved-remote>
• target: <user>@<private-lan-host>

Actual behavior

The connection target is reachable, but authentication fails.

Observed failure:

• Permission denied (publickey,password,keyboard-interactive)

After that, Codex continues to retry the same connection automatically in the background.

From the local SSH debug output, this looks like:

• TCP connection succeeds
• SSH handshake succeeds
• host key is recognized
• authentication does not succeed
• Codex schedules another reconnect attempt

So this does not look like a raw network outage. It looks like an auth-failed saved connection that keeps being retried.

Why this may be partly environment-triggered

This may have been triggered by my local saved remote setup rather than a pure app-created condition.

For example:

• the saved remote target may be stale
• the expected private key may be missing or no longer valid
• the saved connection may no longer be appropriate to auto-start

So I am not claiming the auth failure itself is necessarily a Codex bug.

Why this still seems like a Codex Desktop issue

Even if the saved connection is stale or locally misconfigured, Codex should handle that state more safely.

After repeated auth failures, I would expect the app to:

• stop retrying automatically after a small number of failures
• mark the connection as failed/idle
• require explicit user action before reconnecting
• clearly surface that this is a credential/authentication problem rather than a transient reconnectable state

The likely issue here is not “SSH auth failed”, but rather:

Codex keeps auto-retrying an auth-failed SSH connection instead of backing off cleanly.

Expected behavior

After authentication fails repeatedly, Codex Desktop should do one of the following:

• stop background reconnect attempts
• classify the connection as non-retriable until user action
• require manual reconnect after credentials are fixed
• make the failed-auth state clearly visible to the user

Impact

• persistent background retry noise
• repeated failed SSH attempts
• unnecessary resource usage
• confusing connection state
• possible contribution to app instability over time

Notes

This should likely be tracked separately from any Git workspace / turn-diff / repo-watcher issues.