Skip to main content
These commands control the lifecycle and state of your Claude Code session — clearing or compressing context, resuming previous conversations, monitoring token and cost consumption, and exiting.

/clear

Wipe conversation history and start fresh

/compact

Summarise history into a compact context

/resume

Return to a previous conversation

/context

Visualise current context window fill

/cost

Show session API cost breakdown

/usage

Show plan usage limits

/exit

Quit Claude Code

​
/clear

Clears the entire conversation history and frees up context. Equivalent to starting a brand-new session while staying in the same terminal. Aliases: reset, new
Arguments: none
Non-interactive: no

​
When to use

  • The conversation has drifted off-topic and you want a clean slate.
  • You’ve hit the context limit and /compact is not sufficient.
  • You want to start a completely unrelated task.
/clear permanently discards the current conversation history from the REPL. It is not recoverable after the session ends. Use /compact if you want to retain a summary for continuity.

​
Example

> /clear

  Conversation cleared. Starting fresh.

​
/compact

Compresses the current conversation into a summary and continues with that summary as the new context. This frees up a large portion of the context window while preserving the essential history. Arguments: [optional summarisation instructions]
Non-interactive: yes (can be used in --print mode)
Disabled by: DISABLE_COMPACT=1 environment variable

​
How it works

  1. Claude Code sends the current conversation to the model with an internal summarisation prompt.
  2. The model produces a concise summary capturing key decisions, code changes, and outstanding tasks.
  3. The conversation is replaced with the summary as a system message, freeing context.
  4. If session memory is configured, the summary is also written to the session memory file.

​
Custom summarisation instructions

Pass an optional argument to guide what the summary prioritises: 
> /compact focus on the changes we made to the authentication module
Without arguments, the default summarisation prompt is used, which tries to capture all important context.
/compact is also triggered automatically by Claude Code when the context window is approaching its limit, unless disabled by configuration.

​
Examples

Compact with default instructions: 
> /compact

  Compacting conversation... done.
  Summary written. Context reduced from 87% → 12%.
Compact with custom focus: 
> /compact only keep context about the database schema and migration plan
Compact in non-interactive (print) mode: 
claude --print "/compact"

​
/resume

Opens a session picker that lets you search through and resume previous Claude Code conversations. When resumed, the full conversation history is loaded back into context. Aliases: continue
Arguments: [conversation id or search term]
Non-interactive: no

​
Usage

Run without arguments to open the interactive session browser: 
> /resume
Pass a session ID or search term to jump directly to a match: 
> /resume auth-refactor
> /resume 01J8KZ...

​
Session browser

The browser shows sessions sorted by most recent activity. You can:
  • Type to filter by session content or description
  • Use arrow keys to navigate
  • Press Enter to load the selected session
Sessions are stored in ~/.claude/projects/. Each project directory contains .jsonl transcript files named by session ID.

​
/context

Displays the current context window utilisation as a colour-coded grid. Each cell represents a fraction of the total context window. Arguments: none
Non-interactive: yes (shows raw numbers in non-interactive mode)

​
Example output

> /context

  Context window: ████████████████░░░░░░░░  63% used
  Tokens: 81,920 / 128,000
  ├─ System prompt:   4,200
  ├─ Tool definitions: 3,100
  ├─ Conversation:   74,620
Colours shift from green → yellow → red as the window fills. When above ~80%, consider running /compact or starting a new session with /clear.

​
/cost

Shows the total API cost and time elapsed for the current session, broken down by model and token type. Arguments: none
Non-interactive: yes
Hidden for: claude.ai subscribers (cost is covered by subscription)

​
Example output

> /cost

  Session cost summary
  ─────────────────────────────────────
  Duration:          14m 32s
  Input tokens:      124,510   $0.3735
  Output tokens:      18,204   $0.2731
  Cache write:        41,200   $0.0412
  Cache read:         82,800   $0.0083
  ─────────────────────────────────────
  Total:                       $0.6961
/cost shows the cost for the current session only. Use /insights to analyse costs across multiple historical sessions.

​
/usage

Displays your plan usage against the current billing period’s limits. Available to claude.ai subscribers only. Arguments: none
Availability: claude.ai subscribers

​
Example output

> /usage

  Plan: Pro
  ─────────────────────────────────────────────
  Messages this period:   342 / unlimited
  Opus usage:              28 / 50  ████████░░  56%
  Resets:                 in 18 days

​
/exit

Exits the Claude Code REPL immediately. Any unsaved state (conversation history, background tasks) is cleaned up gracefully. Aliases: quit
Arguments: none

​
Example

> /exit
You can also exit with the standard terminal shortcut Ctrl+D (sends EOF) or Ctrl+C twice in quick succession.
Background tasks started with run_in_background: true are terminated when you exit. If you want them to continue, detach from the session instead of exiting.

​
/export

Exports the current conversation to a file in your chosen format. Useful for sharing transcripts or archiving sessions. Arguments: [path] 
> /export
> /export ~/Desktop/session-transcript.md

​
/rewind

Steps back to an earlier point in the conversation, discarding messages after the selected checkpoint. Arguments: none (opens interactive rewind UI) 
> /rewind
Use arrow keys to select how far back to rewind. Messages after the selected point are removed from context.

Build docs developers (and LLMs) love

Get started for freeTalk to us