lockstep
5 min readNaman Jain

Claude Code plan mode vs auto mode vs accept edits

Claude Code plan mode, auto mode, accept edits, and Manual compared: what each runs without asking, how to switch with Shift+Tab, and when to use which.

You ask Claude Code to rename a field across the API. In Manual mode it stops twenty times to ask about each edit. In auto mode it finishes before you've read the first file it touched. Neither is wrong; they're different modes for different work, and Claude Code has four you'll use day to day.

This guide compares them from Anthropic's permission modes documentation, checked on 16 November 2026.

The short answer

ModeWhat runs without askingUse it for
Manual (default)Reads onlySensitive work, unfamiliar code
Accept edits (acceptEdits)Reads, file edits, and common filesystem commands in your working directoryIterating on code you'll review with git diff
Plan (plan)Reads, plus classifier-approved exploration commands when auto mode is available; no source editsExploring before a large change
Auto (auto)Everything, with a classifier reviewing each actionLong tasks where you trust the direction

Two more exist for narrower cases: dontAsk, which denies anything not pre-approved, for CI, and bypassPermissions, which skips checks and belongs only in isolated containers and VMs.

Switching modes

Press Shift+Tab to cycle. From auto, the first press goes to Manual, then the cycle runs Manual, accept edits, plan. The status bar shows the current mode, such as ⏸ plan mode on or ⏵⏵ auto mode on.

To start in a mode:

claude --permission-mode plan

Or set it per project in .claude/settings.json:

{
  "permissions": {
    "defaultMode": "plan"
  }
}

One exception: "auto" doesn't take effect from a project's .claude/settings.json or .claude/settings.local.json. Put it in ~/.claude/settings.json instead.

Plan mode: think before editing

Plan mode tells Claude to research and propose changes without making them. It reads files, runs commands to explore, and writes a plan, but doesn't edit your source until you approve. Enter it with Shift+Tab, or prefix a single prompt with /plan.

When the plan is ready, you choose:

  • Yes, and use auto mode: approve and let Claude work without routine prompts.
  • Yes, manually approve edits: approve and review each edit.
  • No, keep planning: tell Claude what to change.

Press Ctrl+G to edit the plan in your own editor before approving. At the approval prompt, press Esc to back out; Shift+Tab there approves the plan.

Use it for changes that touch many files, unfamiliar code, or anything where a wrong first step is expensive. Reading a plan takes a minute; unwinding a wrong implementation takes much longer.

Accept edits: fast edits, reviewed later

Accept edits lets Claude create and edit files in your working directory without asking. It also auto-approves mkdir, touch, rm, rmdir, mv, cp, and sed on paths inside that directory.

These still prompt:

  • Paths outside the working directory
  • Writes to protected paths
  • Removals of critical paths
  • Every other shell command, apart from a built-in read-only set

Use it when you'll review the result in your editor or with git diff rather than edit by edit. You still approve each test run or install, which keeps you in the loop on anything that reaches beyond the files.

Auto mode: hands off, with a checker

Auto mode runs without routine prompts. A separate classifier model reviews each action before it runs, and blocks anything that goes beyond your request, targets infrastructure it doesn't recognize, or looks driven by hostile content Claude read. Explicit ask rules still force a prompt.

Blocked by default, among others:

  • Downloading and running code, such as curl | bash
  • Sending sensitive data to external endpoints
  • Production deploys and migrations
  • Force push, and git reset --hard or git clean -fd that would discard uncommitted changes
  • Merging a pull request no human approved
  • terraform destroy and similar

If the classifier blocks 3 actions in a row or 20 in a session, auto mode pauses and Claude Code goes back to prompting. Blocked actions are listed under Recently blocked in /permissions, where you can approve a retry.

From Claude Code v2.1.283, auto mode is the built-in starting mode for interactive terminal and VS Code sessions. It needs a supported model (Sonnet 4.6 or later, or Opus 4.6 or later, on the Anthropic API; older models such as Sonnet 4.5 aren't supported), and an organization can turn it off with disableAutoMode in managed settings.

Anthropic's docs are direct that it reduces prompts but doesn't guarantee safety. Pair it with the sandbox or a container for defense in depth.

Which one to use

  • Starting a big change: plan, then approve into auto or accept edits.
  • Small, clear task in a repo you know: auto.
  • Iterating with tests you want to watch: accept edits.
  • Production systems, secrets, or code you don't know: Manual.
  • CI: dontAsk with an explicit --allowedTools list. See Claude Code headless mode.

Modes set the baseline. Permission rules layer on top, and deny rules block in every mode, including bypassPermissions.

What no mode checks

A mode decides whether Claude asks before acting. None of them check whether the plan matches what your team decided. Auto mode's classifier stops a force push; it doesn't know that the team agreed last week to keep the old endpoint for one more release. A plan can look sensible and still contradict a decision made in a Slack thread Claude never saw.

Where Lockstep fits

Lockstep briefs each Claude Code session on the team's approved decisions that apply to the work, captured from PRDs, Slack, and pull requests, so plans start from what the team agreed. Decisions bind when their owner approves them, and its checks are advisory: they flag a likely conflict for a person to review rather than blocking the change.

See how it works.

Sources checked 16 November 2026

For allow, ask, and deny rules, see Claude Code permissions. For the commands that go with these modes, see the Claude Code slash commands cheat sheet. For workflow advice, see Claude Code best practices.

Frequently asked questions

What is the difference between plan mode and auto mode in Claude Code?
Plan mode researches and writes a plan but doesn't edit your source until you approve it. Auto mode does the work without routine prompts, with a separate classifier model reviewing each action before it runs. A common flow is to plan first, then approve the plan into auto mode.
How do I switch modes in Claude Code?
Press Shift+Tab to cycle. From auto, the first press goes to Manual, then the cycle runs Manual, accept edits, plan. You can also start in a mode with claude --permission-mode plan, or prefix a single prompt with /plan.
What does accept edits mode do?
It lets Claude create and edit files in your working directory without prompting, and auto-approves common filesystem commands such as mkdir, mv, cp, rm, and sed on paths inside it. Other shell commands, protected paths, and paths outside the working directory still prompt.
Is auto mode the default in Claude Code?
From Claude Code v2.1.283, auto mode is the built-in starting mode for interactive terminal and VS Code sessions on every plan and provider. It needs a supported model, and an organization can turn it off in managed settings, in which case sessions start in Manual.
Is auto mode safe?
Anthropic's docs say it reduces prompts but doesn't guarantee safety. The classifier blocks actions such as curl piped to bash, force pushes, production deploys, and sending sensitive data outside, by default. Use it for work where you trust the direction, and keep review for sensitive operations.

Keep reading