A new permission prompt system has been introduced for tool usage, providing better control over what actions Claude can perform.
Key components:
- Tools can now prompt for permission before execution
- Two permission behaviors:
allow(with optional input modifications) ordeny(with explanation) - Permission decisions are tracked with detailed reasoning
Example interaction: When Claude attempts to use a tool that requires permission, you'll see a prompt asking whether to allow the action. You can:
- Allow the tool to proceed (optionally with modified parameters)
- Deny the tool with a reason