Claude Code plan mode: how to start it and what it blocks

In brief

Claude Code plan mode makes Claude read your project and propose changes without editing files until you approve, as of November 2026. Press Shift+Tab until the status bar shows plan mode on, type /plan, or launch with claude --permission-mode plan. Auto mode now starts sessions, so Shift+Tab takes three presses.

Whizi doesn’t connect to Claude Code.

What is the shortcut for Claude Code plan mode?

The shortcut is Shift+Tab, pressed until the status bar under the prompt reads ⏸ plan mode on. Anthropic's permission-modes page, read on 15 November 2026, gives three ways in: that key, a /plan prefix on one prompt, and a launch flag. There is a catch for new sessions. With Claude Code v2.1.283 or later, auto mode is the built-in starting mode in the terminal and VS Code, and the cycle from auto runs auto, Manual, accept edits, plan. That is three presses.

A rolled blueprint resting on a charcoal plinth beside a terracotta mechanical pencil, and a small orange Claude spark sculpture on the next plinth, with a glass toggle switch standing on the right, under a warm rim light.
  1. Install Claude Code and open a terminal in your project folder. How to use Claude Code has the install command for each system.
  2. Run claude, then press Shift+Tab until the status bar shows ⏸ plan mode on. On Windows, Anthropic lists Alt+M as the key when the Node or Bun runtime doesn't enable VT input mode.
  3. Or skip the cycle. Type /plan and Claude Code enters plan mode. Add a task, as in /plan fix the auth bug, and it starts planning that task at once.
  4. Or start there from the shell with claude --permission-mode plan. The same flag works with -p for non-interactive runs.
  5. Describe the change you want and ask Claude to explore first. It reads files, runs exploring commands and writes a plan.
  6. Read the plan, then approve it, edit it or send Claude back to work. The next section covers the approval menu.
From a default session, plan mode is three Shift+Tab presses away
Starting in auto modethe built-in start since v2.1.2833 pressesStarting in Manualauto mode off or an older version2 pressesStarting in accept editsafter one press from Manual1 presseach square is one press of Shift+Tab in the terminal, per Anthropic's docs
Where you run Claude CodeHow to pick plan modeDefault start
Terminal and JetBrains terminalShift+Tab, /plan, or claude --permission-mode planpermissions.defaultMode in a settings file
VS Code extensionClick the mode indicator under the prompt box and pick PlanclaudeCode.initialPermissionMode set to plan
Desktop app, Code tabMode selector in the row under the prompt box; Plan applies to that session onlydefaultMode in settings
claude.ai/codeMode dropdown under the prompt boxCloud sessions offer Accept edits, Plan and Auto
Mobile appTap + in the prompt box, then PermissionSame three cloud modes

Anthropic's desktop guide gives the same advice for all of them: "Start complex tasks in Plan so Claude maps out an approach before making changes." The command reference says /plan "switches into plan mode before a large change."

Does Claude Code plan mode run commands or edit files?

It blocks file edits until you approve, and it lets exploring commands run. Anthropic's permission-modes page puts it this way: "Plan mode tells Claude to research and propose changes without making them." The same paragraph says Claude "runs shell commands to explore", so plan mode is not a read-only sandbox for the shell. What happens to a command depends on where you run it, and Anthropic lists the cases in order.

Where you run itFile editsShell commands while planning
Terminal, auto mode available (the default)Blocked until you approve the planA classifier reviews them and blocks the risky ones; recognized rm critical-path removals still ask you
Terminal, auto mode unavailable or useAutoModeDuringPlan set to falseBlocked until you approveAnything outside the built-in read-only set asks first
Interactive terminal with bypass permissions availableNot enforcedRun without a classifier or prompt; explicit ask rules and rm critical-path removals still prompt
claude -p, Agent SDK, VS Code chat panelBlockedSame rules as the first two rows

The third row is the one most guides miss. Anthropic's page says that in an interactive terminal with bypass permissions available, Claude Code "doesn't enforce plan mode's blocks" and that "a file edit or shell command it attempts during planning runs without prompting." Claude is still told to plan without editing, and explicit ask rules and recognized rm and rmdir critical-path removals still prompt. A told model is not a locked door. If you rely on plan mode for safety, check your launch flags before the first session, and again whenever a teammate changes the shared settings files or the alias you start Claude Code with. Bypass permissions appears when you start with --dangerously-skip-permissions, --allow-dangerously-skip-permissions, or a bypassPermissions default in user, --settings or managed settings.

Plan mode writes one thing to disk, the plan itself, as a file you can open in any editor later, keep next to the code for review, or delete once the work ships. Anthropic's settings reference says plan files go to ~/.claude/plans unless you set plansDirectory, a path relative to the project root, for example "plansDirectory": "./plans". A path outside the project root falls back to the default folder.

How do you review, edit and approve a plan?

When the plan is ready, Claude shows an approval menu with three answers. Anthropic names them: "Yes, and use auto mode", "Yes, manually approve edits", and "No, keep planning". If auto mode isn't available to your session, the first one reads "Yes, auto-accept edits". Approving a plan exits plan mode and starts editing in the mode you chose, so the choice is also your next permission mode.

  • Press Ctrl+G to open the plan in your default text editor and change it before Claude proceeds.
  • Pick "No, keep planning" and type what to change, such as "split step 3 so the migration ships first".
  • Press Esc before you press Shift+Tab at the menu. Anthropic warns that pressing Shift+Tab there approves the plan.
  • Set showClearContextOnPlanAccept to true in settings to add a first option, "Yes, clear context and ...", which approves the plan, clears the conversation and builds from the plan alone.
  • To plan again after approving, press Shift+Tab back to plan mode or start the next prompt with /plan.

The clear-context option matters on big plans. Anthropic's settings reference says "Planning can use a lot of context", and the menu label shows how much of your window the planning took. Accepting a plan also gives the session a title generated from the plan, unless you named the session already.

Prompts that work well here are specific about the outcome and the stop line. Three you can paste: "Read the auth module and the tests, then plan how to add refresh-token rotation. List every file you would touch." Or: "Plan the migration from REST to GraphQL for the orders endpoints in three shippable steps, and name the riskiest step." Or: "What would break if we rename the User table? Plan the safest order of changes."

Claude Code plan mode or auto mode: which should you use?

Use plan mode when the cost of a wrong direction is high, and auto mode when you trust the direction and want no interruptions. Anthropic's cost guide makes the same point in money terms: plan mode before implementation helps by "preventing expensive re-work when the initial direction is wrong." For a one-line fix, a plan is overhead, and Manual or accept edits is quicker. That last rule is our reading, not Anthropic's wording.

ModeWhat runs without askingBest for, per Anthropic
default (Manual)Reads onlyReviewing every action, sensitive work
acceptEditsReads, file edits, mkdir, touch, mv, cp and similarIterating on code you review afterwards
planReads, plus classifier-approved commands when auto mode is availableExploring a codebase before changing it
autoEverything, with background safety checksLong tasks, fewer prompts
dontAskPre-approved tools onlyLocked-down CI and scripts
bypassPermissionsEverythingIsolated containers and VMs only

A common pairing is plan first, then auto. Approve with "Yes, and use auto mode" and Claude builds the plan without asking about each edit. If you want to read every diff instead, choose "Yes, manually approve edits". Team tasks need more care: Anthropic's cost page says agent teams use "approximately 7x more tokens than standard sessions" when teammates run in plan mode, because each teammate keeps its own context window.

Can Opus plan and Sonnet write the code?

Yes. The opusplan model alias uses Opus in plan mode and switches to Sonnet to execute, which Anthropic's model page describes as pairing "Opus's reasoning for planning with Sonnet's efficiency for execution." Set it with /model opusplan inside a session or launch with claude --model opusplan. Anthropic says /model accepts opusplan[1m] from v2.1.265 on, to request the 1M context window for both phases.

Two details from the same page. On current models both phases already run with the 1M window on the Anthropic API. And if your organization restricts models with availableModels, opusplan plans on the newest permitted Opus and stays on Sonnet only when every Opus is excluded. The page also describes a Haiku session upgrading to Sonnet while in plan mode.

Which plan does Claude Code plan mode need, and what does planning cost?

Plan mode needs a paid Claude plan, because Claude Code itself does. Anthropic's pricing page, read on 15 November 2026, marks Claude Code "No" on Free and "Yes" on Pro, and says: "Claude Code is included in all paid plans. It shares the same usage limits as the rest of your plan, so your work in the terminal and your chats draw from one pool." Pro lists at $20 a month, or $17 a month billed as $200 up front. Plan mode has no separate allowance in the docs we read, so planning turns spend the same five-hour and weekly budget as any other turn. Claude Code usage limit by plan shows how that budget is sized, and Claude Code pricing compares every way to pay.

Whizi, which publishes this page, doesn't run Claude Code and has no plan mode, so nothing in the steps above happens there. It helps with one slice: the planning questions that need no files, such as which database fits a prototype or how to split a migration. A Whizi Pro chat puts Claude, GPT and Gemini in one app, so you can ask all three the same design question and paste the best outline into Claude Code as your first message. Pro's first week is $1.99, and the free plan gives 10 messages a day on Whizi AI, Auto and Claude Haiku 5.5, no card.

Why isn't Claude Code plan mode working?

Most plan mode problems come from a handful of documented behaviours. Check these in order.

  • Shift+Tab lands on the wrong mode. The cycle depends on where you started and on which optional modes are enabled. Watch the status bar, not the count. bypassPermissions and auto slot in after plan, and dontAsk never appears in the cycle.
  • Shift+Tab doesn't register on some Windows setups. Use Alt+M when the Node or Bun runtime doesn't enable VT input mode, or skip the key with /plan.
  • Plan mode isn't the default in VS Code. The extension never reads a project's .claude/settings.json for the starting mode. Set claudeCode.initialPermissionMode to plan in your VS Code user settings. Picking Plan from the indicator applies to that conversation only.
  • A project default is ignored. Terminal sessions honour "defaultMode": "plan" in .claude/settings.json. Only auto and bypassPermissions are refused from that file.
  • Edits happened anyway. You may have bypass permissions available, where plan mode's blocks are not enforced. Restart without the bypass flags.
  • Plan mode pauses on every command. With auto mode unavailable, or useAutoModeDuringPlan set to false, commands outside the read-only set ask first. That setting appears in /config as "Use auto mode during plan".
  • /ultraplan is gone. The command reference lists it as "Removed" and points to plan mode. It used to send a planning task to a cloud session.

Anthropic's permission-modes page also notes that picking Plan in the desktop app applies to that session only, and that a mode you pick in a local desktop session is remembered per folder. If a new session starts somewhere unexpected, check the selector before your first message.

Workflow checklist
  • Press Shift+Tab until the status bar shows ⏸ plan mode on, or type /plan, or launch with claude --permission-mode plan
  • Expect three Shift+Tab presses from a fresh auto mode session, and two from Manual
  • Check that bypass permissions is not available if you rely on plan mode to stop edits
  • Press Esc before Shift+Tab at the approval menu, because Shift+Tab there approves the plan
  • Use Ctrl+G to edit the plan in your editor, and set showClearContextOnPlanAccept to build from the plan alone
  • Try /model opusplan to plan on Opus and build on Sonnet
Common questions

Frequently asked questions

What is the shortcut for plan mode in Claude Code?

Shift+Tab, pressed until the status bar shows ⏸ plan mode on. From the default auto mode that is three presses, from Manual two. You can also type /plan or launch with claude --permission-mode plan, and on Windows Anthropic lists Alt+M when the Node or Bun runtime doesn't enable VT input mode.

Does Claude Code plan mode edit files?

No, not before you approve, with one exception. File edits stay blocked until you approve the plan, but in an interactive terminal with bypass permissions available Anthropic says plan mode's blocks are not enforced. Exploring shell commands can run, reviewed by the auto mode classifier by default.

What is the difference between plan mode and accept edits mode?

Plan mode proposes changes and blocks file edits until you approve. Accept edits mode lets Claude edit files and run basic filesystem commands such as mkdir, touch, mv and cp without asking. Shift+Tab goes from Manual to accept edits to plan, so plan mode sits one press after accept edits.

Is plan mode available on the free Claude plan?

No. Plan mode is part of Claude Code, and Anthropic's pricing page lists Claude Code as "No" on Free and included on all paid plans, starting with Pro at $20 a month. Planning draws on the same usage limits as the rest of your plan.

How do I make plan mode the default in Claude Code?

Set "defaultMode": "plan" under permissions in .claude/settings.json for one project, or in ~/.claude/settings.json for every terminal session. The VS Code extension ignores project settings for this, so set claudeCode.initialPermissionMode to plan in your VS Code user settings instead.