IT之家 October 29 news, the tech media outlet The Register published a blog post yesterday (October 28), reporting that in Claude Code version v2.1.311, Anthropic added a new environment variable; after the user enables it, when the AI agent performs operations such as deleting files, it no longer asks the user for approval each time.
The full name of this environment variable is CLAUDE_CODE_EVEN_MORE_DANGEROUSLY_SKIP_PERMISSIONS, literally translated as “skip permission checks even more dangerously.”
Anthropic technical staff member Daisy Hollman explained that the original “dangerously skip permission checks” option does not let all operations through; the system will still intercept certain obviously dangerous commands, such as recursively force-deleting the contents of the user’s home directory; the new setting allows users to further disable this kind of protection. IT之家 attaches the relevant tweet screenshot below:
Hollman said:
Fun fact: --dangerously-skip-permissions actually does not approve all operations; we still block some obvious operations, such as `rm -rf ~` (deleting files).
But today’s version allows you to disable this feature by setting a new environment variable CLAUDE_CODE_EVEN_MORE_DANGEROUSLY_SKIP_PERMISSIONS=1` (if you really need to).
Because the variable name is too blunt, and because the change was not listed in the corresponding version’s changelog, some people doubted whether the news was true. Hollman then clearly responded: “Not a joke.”
However, this does not mean that Claude Code has canceled all safety measures by default. The report points out that the new feature requires users to actively set it, and other restrictions imposed by organization administrators may still take effect.