How to Run Claude Code Entirely in the Cloud: Web Sessions vs. a Remote VM
You can run Claude Code on a remote machine, so your laptop does not need to hold the code, tools, or active…
When you ask Claude Code to change software, Plan Mode lets you review Claude’s proposed approach before it edits source files. It is a planning workflow and a permission mode: it tells Claude to investigate before proposing changes and sets which actions Claude can take without approval.
Use it when you are not sure how much work the task involves, when it touches several parts of a codebase, or when redoing the work would be costly. For a small, clearly defined edit, planning can add an unnecessary step. In both cases, read the plan and check its assumptions. Approving it does not guarantee that Claude will implement the change correctly.
Claude can read project files and explore the codebase. It can also run shell commands for investigation and present a proposed approach. After Claude presents its approach, you can approve the plan. You can also ask Claude to revise it or open the plan in a text editor with Ctrl+G. After you approve the plan, you choose how closely to supervise Claude’s edits, based on the permission settings available in the session. The permission mode documentation describes these options.
In the CLI, enter Plan Mode with Shift+Tab, start a prompt with /plan, or launch Claude Code with claude --permission-mode plan. Other supported interfaces provide their own controls.
The mode is designed to hold off on source edits while Claude develops the plan. Shell commands are a separate consideration: Claude can use them to investigate, and their handling depends on session permissions and configuration. Plan Mode therefore does not act as a general-purpose security sandbox.
Claude bases its plan on the information it can see. If the repository does not reveal a requirement or constraint, Claude may leave it out. Treat the plan as a proposal to inspect, and provide the context and checks needed to judge the eventual implementation. A plan presented during a session should not be assumed to be a durable project specification; when requirements need to persist, record them separately.
Use Plan Mode when Claude must choose between approaches that would lead to different changes. For example, a feature might fit into an existing service or require a new component. You can review the files and approach Claude proposes before it edits, then correct that assumption early.
Use it when Claude must change several files, when you are unfamiliar with the relevant code, or when a wrong approach would take substantial work to undo. Anthropic’s usage guidance suggests considering Plan Mode or a file-by-file outline for work involving more than two or three files. That is a practical rule of thumb, not a cutoff.
Reviewing the plan is what makes planning useful. Check whether Claude has identified the right parts of the codebase, accounted for constraints, and proposed a way to verify the change. If the plan misses an important condition, correct it before approving implementation. For larger tasks, Anthropic’s power-user tips recommend refining the plan before implementation and returning to planning if the work goes off course.
For a precise, low-risk edit in one isolated area, ask Claude to name the files it expects to change and describe each change. That gives you a quick scope check without a full planning round. Anthropic’s usage guidance presents this as an alternative for smaller tasks.
If you have not decided what the software should do, Claude may start planning how to build it before you have defined the requirements. First clarify the requirements. Anthropic’s best-practices guidance recommends having Claude interview you and write a specification when the requirements need to be worked out and preserved. A specification records the intended behavior; a Plan Mode proposal focuses on how Claude might implement a task.
Planning also has a cost in time and tokens. Anthropic describes a plan as costing a few hundred tokens and contrasts that with a hypothetical 400-line change that must be reverted and regenerated, which could cost thousands of tokens plus additional turns. Those figures are examples, not measured averages or proof that planning improves completion time or success rates.
Plan Mode does not verify the code, enforce every project requirement, or guarantee that Claude’s proposal is complete. Run the relevant checks and review the resulting changes after implementation. Planning, testing, code review, and permission controls address different parts of the work.
It also does not guarantee that no command can have side effects before you approve a plan. Command behavior depends on the active permissions and session configuration. For close oversight, check those settings and use stricter action approval or other access controls. Anthropic documents Manual mode separately as an option for reviewing actions; permission settings govern action approval, while Plan Mode provides a checkpoint for reviewing the proposed approach.
Give Vroni a GitHub issue, bug report, spec, or rough idea. It reads the repo, plans the change, writes code, runs checks, and works toward a review-ready pull request.
Take a look at vroni.com