Anthropic has introduced a new environment variable, CLAUDE_CODE_EVEN_MORE_DANGEROUSLY_SKIP_PERMISSIONS, in version 2.1.311 of Claude Code. When set to 1, this variable removes the final safety confirmation prompt for destructive file operations, such as deleting critical system paths.
Previously, the tool included a bypassPermissions mode that removed standard approval prompts but retained a final safeguard for critical deletion commands. The new variable eliminates this last barrier, allowing the AI agent to execute commands like rm and rmdir without human intervention.
The company maintains a five-tier permission system: Manual, acceptEdits, plan, auto, and bypassPermissions. The new variable functions as an additional layer below these tiers. Anthropic advises that such bypass modes should only be utilized within isolated containers or virtual machines to prevent accidental data loss.
Data indicates that users approve approximately 93% of manual permission prompts in Claude Code. Anthropic notes that the explicit naming of the new variable is intended to ensure that developers make a conscious decision before enabling the feature, effectively serving as a liability waiver.
