App review now supports up to 16,384 output tokens, still bounded by the selected model's lower limit.
Responses stopped at the output limit cannot become completed assessments. Required checks and final administrator approval remain mandatory.
Final cross-part review sends unanswered questions instead of repeating completed reports and prioritizes explicitly named source. All existing findings remain preserved.

Update Core and Agent Gateway together. Existing review policies and submission snapshots are unchanged. Administrators can validate and activate a new policy for new submissions when needed. No App code, SDK update or database migration is required.