Back to rankings

Claude Code Fixes Permission Dialog Over-Grant Bugs

Claude CodeView original changelog

Claude Code 2.1.235 patched two permission-dialog bugs that could leave a user granting more access than intended. Pressing Shift+Tab while typing a comment inside a permission prompt was silently approving the pending edit and granting session-wide edit permission, instead of simply closing the comment field. Separately, Anthropic corrected cases where a permission dialog's displayed text and its "don't ask again" option did not accurately reflect the scope of the grant it produced, and the option is now withheld whenever a dialog cannot fully display what is being approved.

Key Takeaways

  • Shift+Tab inside a permission prompt's comment field was silently granting session-wide edit permission instead of just closing the field.
  • The bug could trigger from ordinary text editing, not a deliberate approval click, making it easy to miss.
  • Permission dialogs' displayed text and "don't ask again" option now always match what the grant actually covers.
  • "Don't ask again" is now withheld whenever a dialog can't fully display what's being approved, closing a shown-versus-granted gap.
  • Both fixes address a user's ability to trust that a permission dialog shows exactly what it grants, a core piece of Claude Code's trust boundary.
  • The fixes ship automatically in 2.1.235 with no workflow change required from users.

Two Permission-Dialog Bugs Closed in One Release

Claude Code's permission system is the primary control a user has over what the agent is allowed to touch on their machine, so bugs that cause it to grant more than intended are treated as high priority. Version 2.1.235, released August 18, 2026, closed two such bugs.

Shift+Tab Was Silently Granting Session-Wide Edit Access

Inside a permission prompt's comment field, a text area meant for leaving a note, pressing Shift+Tab was supposed to simply close the field. Instead, it was being interpreted as the keyboard shortcut for "allow all edits during this session," both approving the pending edit and granting the agent blanket edit permission for the rest of the session. Because the bug fired from an ordinary text-editing keystroke rather than a deliberate approval action, a user commenting on a permission prompt could end up granting far more access than intended, without realizing it. Claude Code now closes the comment field on Shift+Tab and leaves the permission decision untouched.

Permission Dialogs Now Match What They Actually Grant

Separately, Anthropic fixed cases where a permission dialog's own display text, and its "don't ask again" checkbox, did not accurately describe the scope of the permission a user was about to grant. The mismatch meant a user could read one thing on screen and unknowingly approve something broader. The fix makes the display text and "don't ask again" option always match what a grant would actually cover, and the "don't ask again" option is now withheld entirely whenever the dialog cannot fully display everything that would be approved, closing the gap between what is shown and what is granted.

Why This Matters

Both fixes fall inside Claude Code's trust boundary: they govern what the agent can do without asking again, and both bugs could previously result in a wider grant than the user consciously chose. Neither requires a workflow change; both fixes land automatically with the update.