A full disk during sign-in refresh is now reported as a full disk

What

Claude Code keeps you signed in by refreshing a login token (an OAuth token) in the background, and saves it in its config folder, ~/.claude by default. When that refresh fails because the disk holding the folder, or your disk quota on it, is full, Claude Code now treats this as its own error:

• The new message says the disk or quota is full, asks you to delete files you no longer need on that disk and try again, and says signing in again won't help.
• A new error type, OAuthRefreshDiskFullError, carries an oauthRefreshDiskFull marker and is classed as oauth_refresh_disk_full, logged at error level next to other sign-in failures.
• When it is retried, the failure is classed as an authentication failure, so it is not retried as if it were a passing glitch.
• Taking the lock that stops two token refreshes running at once now reports lock_disk_full when the disk is full, and lock_error for other failures. A new disk full reason sits beside the existing reasons for clearing credentials.
• Where a session stops over a sign-in error, the reason is now disk_full rather than auth when the disk was full.

Separately, sending usage data on exit now waits for a token refresh in progress by default (waitForOAuthRefresh).

Why

Before, a full disk looked like a login problem, so people could sign in again and again without fixing anything. The message now points at the real cause.