Interactive Mode

How to use Claude Code’s interactive REPL for back-and-forth conversations, plan-based workflows, and real-time cost tracking.

Claude Code runs as an interactive REPL (Read-Eval-Print Loop) — a persistent session where you type requests, Claude responds and acts on your codebase, and the conversation continues until you exit.

Starting a session

Launch Claude Code from your terminal in any directory:

claude

Claude loads any CLAUDE.md memory files it finds in your project hierarchy, then presents an input prompt. You can start typing immediately.

Keyboard shortcuts

Shortcut	Action
Enter	Submit the current input
Ctrl+J	Insert a newline (multiline input)
Ctrl+C	Cancel the current in-progress response
Ctrl+D	Exit the REPL (same as /exit)
Up / Down	Cycle through previous inputs
Ctrl+O	Toggle conversation transcript view

Press Ctrl+C once to interrupt a running response. Claude will stop its current action. You can then type a correction or follow-up.

Input modes

Single-line input

Type your request and press Enter to submit.

> refactor the login function to use async/await

Multiline input

Use Ctrl+J to insert a newline within your message. This lets you provide structured prompts before submitting:

> Fix the following issues:
  1. The database connection is not closed after each query
  2. Error messages are not localized
  (Ctrl+J after each line, Enter to submit)

Pasting file content

You can paste file contents or long blocks of text directly into the prompt. Claude reads pasted content as part of your message.

How Claude reads your project context

When you start a session, Claude automatically loads:

• Project memory — CLAUDE.md and .claude/CLAUDE.md files discovered by walking from your working directory up to the filesystem root
• Local memory — CLAUDE.local.md files for private project-specific instructions (gitignored)
• User memory — ~/.claude/CLAUDE.md for your global personal instructions
• Rules — all .md files in .claude/rules/ directories

Files closer to your current directory are loaded with higher priority. Claude follows the instructions in these files throughout the session.

Claude does not automatically read every file in your project. It reads files on demand as it works — either because you reference them explicitly, or because it decides to read them to understand the codebase.

Plan mode

Plan mode restricts Claude to read-only operations. Claude can analyze code and produce a plan, but it cannot write files, run commands, or make changes until you approve.

Entering plan mode

> /plan

Claude switches to plan mode and confirms:

Enabled plan mode

Working in plan mode

Once in plan mode, describe what you want Claude to do:

> Refactor the authentication module to support OAuth 2.0

Claude reads files, analyzes the codebase, and writes a plan to a file in your project (typically .claude/plan-<slug>.md). It does not make any changes.

Viewing the current plan

Run /plan again while already in plan mode to display the plan Claude has written:

> /plan

This shows the plan content and its file path. If you have an external editor configured, Claude also shows:

"/plan open" to edit this plan in <Editor>

Opening the plan in an editor

> /plan open

This opens the plan file in your configured editor ($VISUAL or $EDITOR).

Executing the plan

Once you are satisfied with the plan, exit plan mode by switching back to normal mode. You can do this by asking Claude to proceed:

> Looks good, go ahead and implement this

Claude exits plan mode and executes the changes described in the plan.

Plan mode uses the same model as normal mode. Switching to a more capable model before entering plan mode will produce higher-quality plans.

Cost and token tracking

Viewing session cost

> /cost

This shows the total API cost and duration for the current session. For Claude.ai subscribers, it shows subscription usage information instead.

Context usage visualization

> /context

Displays a colored grid showing how much of the context window is currently used. This helps you decide when to use /compact to free up space.

Automatic compact warnings

As your conversation grows, Claude Code tracks how much context remains. When context is running low, it displays a warning suggesting you run /compact to summarize the conversation. You can suppress repeated warnings by running /compact directly.

Exiting the REPL

> /exit

Or press Ctrl+D. Both cleanly end the session. Background tasks (if any were started with Ctrl+B) continue running.