Plan mode lets Claude think out loud, read code, and propose a full plan before it makes any changes. It's the deliberate, measure-twice-cut-once mode.

What it does

In plan mode Claude can run read-only tools (Read, Grep, Glob, WebFetch) but is blocked from Edit, Write, and most Bash. The output is a plan - what it would do, in what order, and why. You review the plan, tweak it, then exit plan mode to execute.

When to use it

Large refactors where a wrong first step is expensive.
Exploring a new codebase before touching anything.
Planning a multi-file change you want a human to sign off.
Any task where the cost of a bad edit is higher than the cost of a slower first turn.

Gotchas

Plan mode can still read sensitive files. It's about preventing writes, not reads.
A great plan still needs a careful execution turn. Don't skip the review.
Exiting plan mode without executing is fine - you can use it as pure pre-reading.

Official docs: https://code.claude.com/docs/en/permission-modes.md#analyze-before-you-edit-with-plan-mode