Auto-mode setup in 2.1.268 replaced unrelated user settings, including hooks and permission rules

Summary

In a recorded Claude Code 2.1.268 incident on macOS, saving the result of /auto-mode-setup replaced previously populated user settings with a much smaller configuration instead of preserving unrelated settings.

The resulting configuration retained only permissions, model, and autoMode at the top level. The permission allow list was empty, and the previously configured hook events and other unrelated user settings were absent.

Observed sequence and evidence

The incident record contains the setup-command invocation, followed by its settings-save completion message. The settings file's modification time matched that completion message to the second. The recorded before/after comparison showed the loss of unrelated configuration.

This is a historical incident report, not a new reproduction. The evidence does not establish whether a person or an agent initiated the command, and this report makes no attribution claim.

Expected behavior

Saving setup output should preserve unrelated user settings and hooks. Any proposed removal of existing permission rules should be separately presented for informed acceptance, rather than removing them as an unintended consequence of saving generated configuration.

Impact

User-configured hooks, permission rules, and other preferences were lost from the shared user configuration. This can change the behavior and safety controls of subsequent sessions, not just the session running setup.

Current-version verification status

Not tested end to end on a current release. We do not claim that the defect still reproduces, that it has been fixed, or that an earlier release behaved differently.

The current documentation says setup requires an eligible subscription and feature-flag fetching. We have not established a fully isolated authenticated test environment and have not invoked the command against real user settings as an experiment. Inspecting installed version metadata is not a reproduction test.

Suggested regression coverage

In an isolated maintainer test, seed synthetic user settings containing unrelated top-level fields, hooks, permission entries, and existing auto-mode rules. Exercise the actual setup acceptance/save path and verify that unrelated values remain intact. Test any optional permission-rule cleanup separately, including declining it.

This is proposed coverage, not a procedure we have successfully executed on a current release.

Request

Please confirm whether the setup save path in 2.1.268 could replace rather than merge user settings, whether this has since been corrected, and whether a settings-preservation regression test covers the current implementation.

No original settings, transcripts, credentials, or identifying system details are included in this report.