When something fails, when Claude catches its own mistake, or when a user reports a problem, Claude will now write up a feedback report on its own. Users can review, edit, and approve before it's sent.
The engineer who built it explains that Claude now has a SendFeedback tool, so instead of users invoking /feedback manually and writing a report, Claude can draft the report itself for you to review and approve before sending. The feedback data helps the team understand and fix problems.
The auto-feedback drafting feature can be disabled or adjusted through /config, with full details in the documentation.
Brief aside from the engineer behind the new SendFeedback feature noting that Claude's ability to draft its own feedback reports turned out better than expected.
The Admin API is now accessible through the official SDKs and the ant CLI, covering member management, workspace and API key administration, and org-level rate limit reads.
After a user flagged a problematic example in the Anthropic docs (apparently predating a recent Cursor-related discussion), the docs have been fully updated. A follow-up post about the underlying issue is planned.
