You have just cloned an unfamiliar repository and want to understand how its auth layer works before anything gets changed. Which mode fits?

The principle — when the goal is understanding rather than changing, pick the mode that cannot change anything.

Why the key is correct — Anthropic defines plan mode as a permission mode where Claude researches and proposes changes without editing your source files. That matches the request exactly: you want the auth layer explained and an approach proposed, with nothing landing on disk while you are still forming a view of the codebase. Reading and searching stay available, which is what makes the exploration useful, and the write is the only thing withheld.

Why the others are wrong — approving edits one at a time in the ordinary mode still produces edits, which is a different activity from understanding first; careful changing is not the same as not changing. Writing a deny rule for the auth folder leaves the rest of an unfamiliar repository writable, and the point of exploring is that you do not yet know which other parts the work would touch. And a mode does not enumerate commands at all — that is what rules express — so reaching for a list is solving this at the wrong layer.

Remember this — in a repository you do not know yet, plan first and let the plan be the thing you review.

Sources — Anthropic's Claude Code glossary.