What is plan mode in Claude Code?
Plan mode tells Claude Code to research and propose changes without making them. It reads files, runs shell commands to explore, and writes a plan — but does not edit your source, and edits stay blocked until you approve the plan. It's one of the permission modes you cycle with Shift+Tab, and the status bar shows ⏸ plan mode on while it's active.
- What it does
- Proposes reads and explores, no edits
- Enter with
- Shift+Tab cycles the modes
- Single prompt
- /plan prefix one instruction
- Status badge
- ⏸ plan mode on shown in the status bar
- Edits
- Blocked until you approve the plan
- Also settable
- By default in your settings file
A permission mode, not a personality setting.
The documentation is unambiguous about the boundary: "Plan mode tells Claude to research and propose changes without making them. Claude reads files, runs shell commands to explore, and writes a plan, but does not edit your source." Edits stay blocked until you approve, except in sessions running with bypass permissions.
Entering it takes one keystroke — Shift+Tab cycles default → acceptEdits → plan — or you can prefix a single instruction with /plan when you only want one answer planned rather than a whole session. It can also be set as the default mode for new sessions in your settings.
Exploration still runs commands, and how those are approved has moved between versions — the doc carries version markers on that behaviour. We've left the specifics out deliberately: they'd be stale within weeks, and the durable fact is the one that matters. Your source doesn't change while you're in plan mode.
When to use it.
The natural use is the start of a big task in a codebase the agent hasn't seen. Let it read, explore and produce a plan; read the plan; correct the misunderstanding that would otherwise have become fifteen wrong edits. Approving a plan is much cheaper than reviewing a diff you didn't expect.
It's also the right mode for asking questions about a repository — the answer needs reading, not writing.
Plan mode constrains effects, not correctness. A confidently wrong plan is still wrong; it's just cheap to reject. Read the plan for the assumption you didn't state rather than skimming it for tone.
And note the exception in the docs: sessions with bypass permissions available behave differently. If someone has configured that, the guarantee you think you have is weaker than the badge suggests.
The cycle runs from most cautious to least: default asks before each change, acceptEdits auto-approves file edits, and plan blocks them entirely. Beyond the cycle sit modes that reduce prompting further, up to a bypass flag whose name is its own warning.
That progression is the practical answer to "is it autonomous" — you choose how much rope, per session.
The other levers in a session.
Plan mode is one of a small set of controls worth knowing before you point the agent at something important.
Frequently asked.
Quick follow-ups people search after this question.
Q · 01 How do I turn on plan mode? +
Shift+Tab to cycle the permission modes until the status bar reads ⏸ plan mode on. For a single instruction, prefix it with /plan instead. It can also be made the default for new sessions in your settings file.