Plan mode
Plan mode is a read-only operational state in Claude Code that separates exploration and planning from execution. When active, Claude can read files, search the codebase, and answer questions, but cannot write files, execute shell commands, or make any changes to your project. This lets you review and refine Claude's proposed approach before a single line of code is touched.
The mode is designed to reduce risk on multi-file changes, refactors, or unfamiliar codebases. Instead of watching Claude cascade edits across a dozen files based on a misunderstood requirement, you get a structured, numbered plan you can audit, amend, and approve first.
Once you are satisfied with the plan, you exit plan mode and Claude executes against the agreed strategy. If you want to adjust the plan before execution, you can open it in your external editor (Ctrl+G), edit the Markdown directly, save it, and Claude will incorporate your changes before proceeding.
When you’d use it
- ◆Reviewing multi-file changes before execution — A developer needs to add a new feature that touches authentication, routing, and database schema files. Before allowing any edits, they enter plan mode to see exactly which files will change and what each change will do.
- ◆Exploring an unfamiliar or legacy codebase — An engineer joins a project with no documentation. They use plan mode to ask Claude to trace data flow and map dependencies across the codebase, knowing Claude cannot accidentally overwrite files or trigger build scripts during exploration.
- ◆Planning complex refactors with uncertain scope — A team wants to migrate synchronous I/O to async across a data layer. The full scope is unknown. Plan mode forces Claude to enumerate all affected modules, dependency chains, and test gaps before touching a single file.
- ◆Enforcing architectural constraints before implementation — A senior engineer has a detailed CLAUDE.md file with project-specific CQRS patterns and integration testing rules. They use plan mode to verify that Claude's proposed implementation respects those constraints before authorizing execution.
- ◆Collaborating on a plan with a team before merging — A developer generates a plan in plan mode, opens it with Ctrl+G, edits it to remove an overly aggressive file deletion step, adds a requirement for E2E tests, and shares the revised plan with teammates for review before any code is written.
What changed recently
- ◆2026-04 — Effort level defaults were adjusted and then reversed. On April 7, 2026, Anthropic reversed an earlier reduction in effort levels. All users now default to 'xhigh' effort for Opus 4.7 and 'high' effort for all other models. This affects the depth of reasoning Claude applies during plan generation.
- ◆2026-04 — The /ultrareview slash command was introduced, producing a dedicated review session that reads through changes and flags bugs and design issues. Pro and Max Claude Code users received three free ultrareviews on launch.
- ◆2026-04-10 — A bug introduced on March 26, 2026 — where Claude's older thinking was being cleared every turn instead of only once after an hour of session inactivity — was fixed. Context management in long plan-and-execute sessions is now stable again.
- ◆2026-03-26 — Claude Code shipped a context management change to clear older thinking from sessions idle for over an hour. A regression caused it to clear every turn; this was subsequently fixed on April 10.
This is the short version
The full chapter has three worked examples, the common pitfalls, and the workflow that makes it pay — plus the other 84 features, kept current.
Get Claude Master — $97 →