Anthropic adds a Claude Code switch that lets agents delete files without asking

Anthropic shipped a new environment variable in Claude Code v2.1.311 that disables the last safety prompt in the agent, including the one that stops it from running rm. The variable name spells out exactly what it does. Anthropic has shipped a way for Claude Code to run destructive shell commands without asking anyone. A new environment variable, CLAUDE_CODE_EVEN_MORE_DANGEROUSLY_SKIP_PERMISSIONS=1, tells the coding agent to stop prompting for approval before executing commands that delete files, such as rm and rmdir. The company introduced the switch in Claude Code build v2.1.311. The change lands in a code assistant that Anthropic markets to enterprises.

First reported by news.lavx.hu. This page is a short excerpt; the full text is theirs.