Claude Code auto mode: a safer way to skip permissions

Anthropic introduces "auto mode" for Claude Code, a model-based classifier system that sits between manual permission prompts and the unsafe `--dangerously-skip-permissions` flag. The classifiers evaluate each tool call at runtime to decide whether it needs human review, based on reversibility, scope, and risk. Anthropic's internal incident log — covering cases like accidental remote branch deletion and auth token exfiltration — motivated the design.