What plan mode is

Plan mode changes the deal: Claude explores your code read-only and hands you a written plan before it touches a single file. It reads, it thinks, it proposes the steps it would take. It does not edit, run mutating commands, or create anything until you approve the plan.

The flow

It is a two-step loop instead of one:

1. Claude investigates and drafts a plan of the changes it intends to make.
2. You read the plan. If it looks right, you approve and Claude executes it. If not, you correct course before any code has changed.

You get a look under the hood before the work happens, which is exactly when a mistake is cheapest to catch.

How to enter it

Headless, pass the permission mode on the command line:

claude -p --permission-mode plan "Add retry logic to the API client"

Claude responds with a plan instead of edits. In the interactive console, use the mode toggle: press Shift+Tab to cycle into plan mode, then prompt as normal and review the plan it returns.

When it earns its keep

Plan mode is not for every one-line tweak. It pays off when the cost of a wrong move is high:

• Unfamiliar code. You are not sure how the pieces connect, so you want Claude's read of it before it starts editing.
• Risky changes. Touching auth, migrations, billing, anything where a bad edit is expensive to undo.
• Large changes. A refactor spanning many files, where a plan lets you catch a wrong assumption before it multiplies across the codebase.

For a quick, well-understood edit, plan mode just adds a step. For anything big, risky, or unfamiliar, reading the plan first is the cheap insurance.

Your lab setup

This VM comes with Claude Code preinstalled and signed in - just run claude to start.