What Claude Code Auto Mode Is and How to Use It

If Claude Code keeps stopping to ask for permission, Auto mode is the feature meant to reduce that friction.
It is not a “run everything without checks” mode.

Anthropic introduced Auto mode on March 24, 2026 as a new permission mode for Claude Code.
The idea is simple: let Claude continue longer tasks with fewer interruptions while still running background safety checks.

This article explains what Auto mode is, how to start using it, and when it makes sense.
It also includes copy-paste prompts built around the product URL so a reader can immediately imagine how to use it.

What Auto mode does

Auto mode is a middle path between two extremes.

In the default mode, Claude Code asks for approval frequently, especially for file writes and shell commands.
That is safer, but it breaks the flow of long-running tasks.

At the other extreme, --dangerously-skip-permissions skips checks entirely.
Anthropic warns that this should only be used in isolated environments.

Auto mode sits between those options.
Safe-looking actions can proceed, risky ones can be blocked, and Claude can try another approach before asking you to step in.

How the safety check works

According to Claude Code Docs, Auto mode evaluates actions in a fixed order:

Existing allow and deny rules are checked first
Read-only actions and file edits in the working directory are approved automatically
Other actions are sent to a classifier
If the classifier blocks something, Claude tries an alternative

Anthropic says the default blocking behavior is designed to catch things like production deploys, destructive source control actions, mass deletion, and suspicious external access.

When Auto mode is useful

Auto mode fits work where you trust the overall direction but want fewer interruptions:

Fixing test failures across several files
Running routine refactors
Installing expected dependencies
Handling repetitive setup or maintenance tasks

When to stay careful

Auto mode is still not a replacement for human review on sensitive work.

It is a poor fit for:

Production deploys
Migrations with hard-to-reverse impact
Secrets handling
Shared infrastructure changes

Anthropic explicitly says Auto mode reduces risk compared with bypassing permissions, but does not eliminate risk.

How to start using it

CLI

claude --enable-auto-mode

Then switch permission modes in the session with Shift+Tab.

Desktop and VS Code

Turn Auto mode on in Settings -> Claude Code, then choose it from the permission mode selector inside a session.

Availability note

As of March 24, 2026, Anthropic announced Auto mode as a research preview for Team users.
The docs also say Team and Enterprise may require admin enablement, and support depends on Claude Sonnet 4.6 or Opus 4.6.

Copy-paste prompts

1. Understand the feature fast

Read this URL and explain what Claude Code Auto mode is for a beginner.
Keep it short, in 3 points, and write in Japanese.
https://claude.com/product/claude-code#auto-mode

2. Decide whether it fits a task

Read this URL and tell me whether Claude Code Auto mode fits my current task.
My task is light edits across multiple files plus test runs.
Separate what is suitable from what I should still review myself.
https://claude.com/product/claude-code#auto-mode

3. Avoid unsafe usage

Read this URL and organize what Claude Code Auto mode is okay for and what I should still approve manually.
Assume I do blog operations and light code changes.
Explain it in simple Japanese.
https://claude.com/product/claude-code#auto-mode

4. Ask for setup steps too

Read this URL and explain how to start using Claude Code Auto mode.
Please organize the answer in the order of CLI, Desktop, and VS Code.
https://claude.com/product/claude-code#auto-mode