Some commands too complex to analyse no longer always ask for permission

Before running a shell command, Claude Code breaks it down to decide whether it is allowed. When a command is too complex to break down, it used to always ask you first.

For a command that has a keyword after its prefix and no other restrictions, Claude Code now does one of these before falling back to asking:

• checks whether the command asked to run outside the sandbox (the boundary limiting what commands can touch) with dangerouslyDisableSandbox
• passes the command on to the normal permission check against your rules