Claude Code Auto Mode vs Accept Edits: Which To Use

Comparing claude code auto mode vs accept edits? I’ve run both modes daily across client sites and my own agency projects, and the honest answer is: they solve two completely different problems. Auto mode hands approvals to a background classifier so Claude keeps working hands-off; accept edits only auto-approves file changes and still asks you about everything else. Here’s exactly what each one does, straight from the official docs, and which one I’d pick for your workflow.

Claude code auto mode vs accept edits: the core difference

Both are official permission modes in Claude Code, and the distinction matters more than most tutorials let on. Everything below comes from the official Claude Code permission-modes documentation at code.claude.com, checked on 22 September 2026.

Auto mode replaces you as the approver. A separate classifier model reviews actions before they run, “blocking anything that escalates beyond your request, targets unrecognized infrastructure, or appears driven by hostile content Claude read” — that’s the docs’ own wording. It covers shell commands, network access, the lot. Auto mode also nudges Claude to keep working instead of stopping for clarifying questions, which is exactly what you want on long agentic runs.

Accept edits keeps you as the approver for anything interesting. It auto-approves file edits and a short list of filesystem commands (mkdir, touch, rm, rmdir, mv, cp, sed) — and only for paths inside your working directory or additionalDirectories. Every other Bash command, any out-of-scope path, and writes to protected paths still throw a prompt. The status bar shows ⏵⏵ accept edits on so you always know where you stand.

What each mode auto-approves (comparison table)

Behaviour	Auto mode	Accept edits
File edits in working directory	Yes (classifier-reviewed)	Yes
Filesystem commands (mkdir, mv, cp, sed…)	Yes (classifier-reviewed)	Yes, in-scope paths only
Other Bash commands (npm, git, curl…)	Yes, if the classifier approves	No — prompts you
Force push, git reset --hard, curl | bash	Blocked by classifier by default	Prompts you
Clarifying questions	Discouraged — Claude keeps working	Normal behaviour
Best for	Long tasks, reducing prompt fatigue	Iterating on code you’re reviewing

One nuance worth knowing: explicit ask rules you set in /permissions force a prompt in both modes, and deny rules block in every mode. Modes set the baseline; rules layer on top.

How to switch modes (and the defaults that changed)

In the CLI, press Shift+Tab to cycle. From auto mode, the first press switches to Manual (default), then the cycle runs default → acceptEdits → plan → back to default, with optional modes slotting in after plan. Or start a session directly in the mode you want:

claude --permission-mode acceptEdits
claude --permission-mode auto

Since the permission-mode docs were updated for v2.1.228+, the built-in starting mode on Pro, Max and Team plans is auto mode — Enterprise plans and API-key sessions still start in Manual. There’s also a neat shortcut from v2.1.247: when a Bash permission prompt appears in Manual or acceptEdits mode, Claude Code adds a “Yes, and switch to auto mode” option, so you can approve the command and go hands-off in one keystroke.

Two gotchas I see people hit constantly:

• Setting "defaultMode": "auto" in a project’s .claude/settings.json does nothing — the value only takes effect from user settings (~/.claude/settings.json) or managed settings.
• Auto mode needs a supported model (Opus 4.6+, Sonnet 4.6+, or a Fable model on the Anthropic API), and organisations can disable it entirely with disableAutoMode. If auto is missing from your Shift+Tab cycle, that’s usually why.

If you’re tuning the rest of your setup at the same time, my guides on the Claude Code effort level and the new Claude Code diff panel pair naturally with this — the diff panel is especially handy in accept edits mode, since reviewing after the fact is the whole point.

Which one should you actually use?

Here’s my honest take after months of running both across four WordPress properties and a stack of client builds:

• Use auto mode for long autonomous tasks: content pipelines, refactors, research runs, anything where prompt fatigue would make you rubber-stamp approvals anyway. The classifier is genuinely conservative — it blocks force pushes, git reset --hard, mass deletions and secret-leaking commits by default.
• Use accept edits when you’re actively pair-coding: you want file saves to flow without interruption, but you still want to see every git and npm command before it runs. It’s the “review changes in your editor or via git diff after the fact” mode, as the docs put it.
• Use Manual for anything sensitive or unfamiliar. The docs themselves warn that auto mode “does not guarantee safety”.

If Claude Code is part of how you do SEO — and it should be — this decision compounds. The right mode is the difference between an agent that ships 20 pages while you sleep and one that sits waiting on a permission prompt at 2am. That’s precisely the kind of pipeline we build together in the AI Profit Boardroom, and if you want an expert pair of eyes on yours, book a free SEO strategy session.

Related setup reading: Claude Code now reads AGENTS.md files natively, which changes how you brief agents before they ever hit a permission prompt.

The bottom line on claude code auto mode vs accept edits

They’re not competitors — they’re different trust levels. Accept edits trusts Claude with your files but not your shell; auto mode trusts a classifier with both. On Pro, Max and Team plans you’re already starting in auto mode, so the real question is when to step down: press Shift+Tab into accept edits when you want eyes on commands, and into Manual when the stakes are high. Set your default once in ~/.claude/settings.json, layer deny rules for the things you never want to happen, and let the mode do the rest.

FAQ: claude code auto mode vs accept edits

Is auto mode the default in Claude Code?

On Pro, Max and Team plans, yes — auto mode is the built-in starting permission mode for new sessions (requires v2.1.228+, per the official docs, September 2026). Enterprise plans and Claude Console API keys start in Manual mode.

What does accept edits mode auto-approve?

File creation and edits in your working directory, plus mkdir, touch, rm, rmdir, mv, cp and sed on in-scope paths. Everything else — other Bash commands, out-of-scope paths, protected paths — still prompts.

What does the auto mode classifier block?

By default: downloading and executing code (curl | bash), sending sensitive data externally, production deploys and migrations, mass deletion on cloud storage, force pushes, destructive git commands like git reset --hard and git clean -fd, and destroying files that existed before the session.

How do I switch between auto mode and accept edits?

Shift+Tab cycles modes in the CLI (from auto: default → acceptEdits → plan). Or start with claude --permission-mode acceptEdits. In VS Code and the Desktop app, use the mode indicator instead.

Why doesn’t defaultMode “auto” work in my project settings?

By design, "auto" is ignored in .claude/settings.json and .claude/settings.local.json. Put it in your user settings file, ~/.claude/settings.json, and it will take effect.

Is auto mode safe?

Safer than blanket bypass, not a guarantee — the docs say exactly that. Deny rules still block in every mode, ask rules still force prompts, and writes to protected paths like .git are never auto-approved outside bypass mode.

Related reading

• Claude Code AGENTS.md Support: What Changed
• Claude Code Effort Level: Full Guide
• Claude Code Diff Panel: See Every Edit Live

Last updated September 2026. This is the living guide to claude code auto mode vs accept edits — it gets updated as the tools change.