You'll noticeTier: how much it should matter to you
1Useful: my rating, 1 to 5
1Signal: worth watching, 1 to 5
PermissionsArea: what it touches
ImprovementsKind: in v2.1.281,
ImprovementsSection of the release
What
A dangerous removal can be denied because nobody answered the approval prompt, or because the session cannot show a prompt at all. In those cases the model now receives a dedicated message. The model is the AI running the commands. The message says:
The denial came from a built-in Claude Code safety check, not from the user.
The model should not get around it by splitting the command, putting it in a script or running it again.
If it applies, the prompt has already gone unanswered a given number of times this session and was not shown again.
Why
Claude no longer mistakes an automatic block for your decision. It is also told not to find a workaround for a removal that the safety check stopped.
