Prefer auto mode over --dangerously-skip-permissions

The problem

Long tasks drown you in approval prompts, so people reach for `--dangerously-skip-permissions`, which can delete, exfiltrate, or run dangerous commands unsupervised.

The hack

Start with `claude --enable-auto-mode`, then Shift+Tab into auto mode. A classifier reviews each tool call; safe actions proceed, risky ones are blocked or escalated.

Why it works

Anthropic positions auto mode as the supported middle path: long-running autonomy with safeguards, now generally available, instead of a blanket permission skip.

Setup

1. Launch with `claude --enable-auto-mode` (or enable in Desktop/VS Code Claude Code settings). 2. Cycle permission modes with Shift+Tab until auto mode is active. 3. Kick off a long task and watch safe steps proceed without prompts. 4. Keep isolated environments for unattended runs — auto mode reduces risk, it does not eliminate it. 5. Admins can disable via managed `disableAutoMode` if needed.

Source

Related

• → Allowlist safe Bash commands to stop approval fatigue
• → Enable Bash sandboxing to auto-allow contained commands