GitHub issue classification now waits through model provider outages instead of immediately failing jobs. When the provider returns a billing, spending-limit, suspension, or rate-limit error, the classification job is snoozed for an hour and retried later, so issues still get classified once the provider account recovers. Provider errors are also sanitized before they reach logs or background job errors, so prompt and request body content no longer leaks to Sentry or Oban. The revision-summary worker uses the same sanitized error handling.
Hive
Snooze and sanitize model provider errors in GitHub issue classification
Published
Jun 29, 2026 · 14:27 UTC
Repository
tuist/hive