Claude Code System Prompt Update

Through the latest update to Claude Code (CC), the system prompt related to agent security monitoring features and tool-use metadata has been changed.

Key update details:

• Enhanced security monitoring (v2.1.179):

  • Clarified the permission scheme so that, when read-only permission has been granted for a specific target, additional read commands for that target execute without requiring separate re-approval.
  • Improved so that when a user re-confirms (e.g., "proceed", "do it"), the specific target and reason for a previously blocked action are inherited, enabling accurate execution of the task without requiring the user to re-enter input.
  • Added security logic related to the Live-Shared Artifact feature: if a shared artifact contains sensitive information (passwords, personal information, etc.), exposure is restricted depending on whether the user is aware of it.

• Introduction of tool-use metadata (v2.1.181):

  • Support added for passing metadata such as display name (displayname), server display name (serverdisplayname), and icon URL (iconurl) during tool use via the toolusemeta field.