← Retour à la liste

Claude Code Command Roundup — Slash Commands, Shortcuts, CLI

claude-codeindie-dev

A cheat sheet of Claude Code commands, organized for my own use. Nobody memorizes all of them, so this is sorted "most used first." (As of July 2026. /help, or typing / in the input box, shows the list for your current version.)

Learn these first

Action What happens
/clear Reset the conversation and start a new session (history is recoverable via /resume)
/compact Summarize the conversation to compress context. /compact instructions also steers the summary
/context Show a breakdown of context usage. When it balloons, time for /compact
/resume Pick a past session from a list and resume it
Esc Interrupt Claude mid-run (work so far is kept). For course corrections
Esc ×2 With empty input: the rewind menu (roll code and conversation back to a checkpoint). With input: clears the draft
Shift+Tab Cycle permission modes (manual approval → auto-approve edits → plan mode…)
!command Shell mode. Run a command directly, bypassing Claude, with output captured into the conversation
@path File-path mention with completion. Point instructions at a specific file

Session and conversation controls

Command Description
/clear Start a fresh conversation (memory like CLAUDE.md is retained)
/compact Summarize and compress the conversation
/context Visualize context usage
/resume Return to a past session
/rewind Roll code and conversation back to a checkpoint (same as Esc ×2)
/fork Copy the current conversation into a separate background session
/btw question Ask a side question without polluting history; the answer shows in an overlay
/copy Copy the last response to the clipboard
/export Export the conversation to a text file
/recap Generate a summary of the session so far

/btw is a sort of reverse subagent: "sees all current context, can't use tools." Great for asking "what was that filename again" without stopping the work.

Settings and environment

Command Description
/config Adjust settings (editor mode = vim, theme, auto-responses, etc.)
/model Switch models
/effort Change thinking depth (low–max)
/fast Toggle fast mode (faster output)
/permissions Configure tool-approval rules
/memory Open and edit CLAUDE.md (memory files)
/init Analyze the project and auto-generate CLAUDE.md
/add-dir path Add a working directory (for touching things outside the repo)
/mcp Manage MCP server connections and auth
/agents Create and manage subagents
/skills List and manage available skills
/hooks Configure pre/post tool-execution hooks
/statusline Configure the status line at the bottom of the terminal
/terminal-setup Set up the Shift+Enter newline keybinding in your terminal
/theme Change the color theme
/doctor Diagnose installation and settings
/login /logout Switch accounts
/usage (/cost) Show token usage and remaining plan quota

Development workflow (bundled skills)

Command Description
/plan Switch to plan mode (have it plan before implementing, then approve)
/code-review Review the current diff. /code-review ultra runs the heavy multi-agent review
/security-review Check the diff for security vulnerabilities
/simplify Review proposing behavior-preserving cleanups only
/review PR# Review a GitHub PR
/deep-research query Search the web and produce a cited report
/loop Run a prompt repeatedly (scheduled / self-driving)
/goal condition Set a goal condition and keep working until it is met
/subtask instructions Delegate a task to a subagent
/tasks List running background tasks and subagents
/background Turn the current session into a background agent

Keyboard shortcuts (the useful ones)

Key Action
Esc Interrupt / close dialogs
Esc ×2 Rewind menu (when input is empty)
Shift+Tab Cycle permission modes
Ctrl+C Interrupt. Clears input if any; press again to exit
Ctrl+O Toggle the transcript viewer (detailed tool-execution log)
Ctrl+R Reverse-search past prompts
Ctrl+B Send the running command or agent to the background
Ctrl+T Toggle Claude's task checklist view
Ctrl+S Stash the text being typed (press again to restore)
Ctrl+G Edit the input in an external editor
Ctrl+L Redraw the screen (when the display corrupts)
Alt+P Switch model (without losing your input)
Alt+O Toggle fast mode
? (empty input) Show the shortcut help
Tab Accept the grayed-out prompt suggestion

Newlines: Shift+Enter (works out of the box in Windows Terminal; VS Code terminals need /terminal-setup), Ctrl+J, or \ + Enter.

Input prefixes

First character Meaning
/ Invoke a command or skill
! Shell mode (run directly; output enters the conversation)
@ File-path completion
? Shortcut help (empty input only)

CLI commands from the terminal

Command Description
claude Start an interactive session
claude "prompt" Start with an initial prompt
claude -p "prompt" Run once non-interactively and exit (for scripts; accepts piped input)
claude -c Continue this directory's most recent session
claude -r name Resume a session by name or ID
claude --model NAME Launch with a specific model
claude update Update Claude Code itself
claude doctor Diagnose the environment
claude mcp Configure MCP servers
claude agents Manage parallel sessions in the agent view

Building custom commands (skills)

Your own slash commands are built as skills. Write the procedure in .claude/skills/COMMAND/SKILL.md in the repository and it becomes callable as /COMMAND — and committing it with the repo shares it with your team (and future you).

.claude/
  skills/
    commit/
      SKILL.md   ← callable as /commit

In my repo, /commit (commit conventions), /blog-article (blog post generation), and /x-announce (X announcement assets) are all skills, which killed the need to retype the same instructions. A good rule of thumb: the second time you paste the same instructions, make it a skill.

Summary

  • Daily life is mostly /clear /compact /context /resume + Esc Shift+Tab ! @
  • When things go wrong, Esc ×2 to rewind. To lock in a design first, /plan
  • /code-review before committing; extract routine work into .claude/skills/
  • Type / in the input box for the current full list. This article is a map of the frequently used ones