On October 29, according to tech media outlet The Register, Anthropic recently quietly added a brand-new environment variable in its release of Claude Code v2.1.311. By enabling this setting, users can allow the AI agent to no longer seek user approval each time it performs high-risk operations such as deleting files.
The full name of this new environment variable is CLAUDE_CODE_EVEN_MORE_DANGEROUSLY_SKIP_PERMISSIONS, which literally translates to “even more dangerously skip permission checks.”
Anthropic engineer Daisy Hollman explained the origin of this feature in a later response. The --dangerously-skip-permissions option built into previous versions, although it sounds like it grants enormous permissions, did not actually fully clear all high-risk operations. The underlying system would still forcibly block some obviously dangerous commands, such as rm -rf ~, which recursively force-deletes the contents of the user's home directory. The extreme environment variable newly added this time is specifically intended to meet the extremely high autonomy needs of some core developers, allowing them to completely disable this last line of protection.
In response to the various questions sparked by the variable's overly blunt and slightly tongue-in-cheek name, and by the fact that it was not listed in the official changelog for the corresponding version, Hollman later publicly confirmed on social media: “This is not a joke.”
However, it should be pointed out that this feature will not be enabled by default for all users. Developers must actively configure it in the environment for it to take effect, and the related security policies and restrictions previously imposed by organization administrators will also continue to apply. For senior developers who pursue ultimate efficiency and want to give AI agents full autonomous operating permissions, this update undoubtedly greatly reduces the cumbersome pop-up confirmation process; but from the perspective of system security, how to strike a balance between fully delegating authority and preventing accidental operations will also become a new challenge that developers need to carefully weigh in actual use.