🤖

Auto Permission Mode

Claude Code

Auto-judge tool usage for long autonomous work sessions

Auto Permission Mode is a mode where Claude Code judges tool usage by itself. Normally, human approval is required for every file edit or command execution, but in auto mode, Claude judges safety and proceeds automatically.

How to Run

claude --permission-mode auto

Advantages

✓
Long-running tasks — Handle complex tasks all at once without waiting for approvals
✓
Autonomous operation — Claude directly decides and executes file reads/writes, test runs, builds, and more
✓
Optimal for batch tasks — Bulk file edits, refactoring, test generation, and more

--permission-mode vs --dangerously-skip-permissions

--permission-mode auto lets Claude judge the safety of tool usage and decide whether to allow or deny it. If deemed dangerous, it will still request approval — safety guardrails remain in place.

--dangerously-skip-permissions does exactly what its name implies — it completely bypasses all permission checks. Any tool will be executed unconditionally, exposing you to risks such as prompt injection. It is recommended only for automated environments like CI/CD.

Mode	Description	Safety
`default`	Human approval required every time	Safest
`plan`	Read tools are automatic, write tools require approval	Safe
`auto`	Claude judges safety and proceeds automatically	Recommended
`--dangerously-skip-permissions`	Ignores all permission checks	CI/CD only

Cautions

⚠Recommended for use only in trusted projects
⚠Using under Git version control makes it easy to review and revert changes
⚠You can switch back to previous behavior with --permission-mode plan or --permission-mode default

How to Use

1

Run with claude --permission-mode auto

2

Claude auto-judges tool usage — permission prompts drastically reduced

3

Continue complex work autonomously for long periods without interruption

4

Use under Git management, and review changes with git diff after completion

Source

Anthropic