Complete Command Cheatsheet

Most people use about 10% of Claude Code.

They type prompts and hit Enter. That's it. They don't know /compact exists. They don't know you can switch models mid-conversation. They definitely don't know that pressing Escape cancels a running operation without losing any context.

This is everything, in one place.

Slash Commands

Type these directly into the Claude Code prompt.

Session Management

Configuration

Workflow

Custom Skills

Any .md file in ~/.claude/skills/ or your project's .claude/skills/ becomes a slash command:

/my-skill        # Runs the skill defined in my-skill.md
/commit          # If you have a commit skill
/review-pr 123   # Skills can take arguments

Keyboard Shortcuts

The Essentials

Power User Shortcuts

Three files in, you see it going wrong. Press Escape to stop (edits so far stay), Shift+Tab into plan mode, then the up arrow to send the same prompt again.

CLI Flags

These go on the claude command when you start a session.

Starting Sessions

claude                       # Start interactive session in current directory
claude "prompt here"         # Start with an initial prompt
claude -p "prompt"           # Pipe mode (non-interactive, for scripts/CI)
claude -c "continue"         # Continue the most recent session
claude --resume              # Resume last conversation (pick up where you left off)

Model & Behavior

claude --model opus          # Start with Opus 5.5 (default)
claude --model sonnet        # Start with Sonnet 5.5
claude --model haiku         # Start with Haiku 5.5
claude --verbose             # Show tool calls and internal reasoning
claude --bare                # Minimal output (for scripted calls)

Context & Files

claude --append-system-prompt "..."  # Add to the system prompt
claude @file.ts              # Load a file into context at start

The --resume flag is criminally underrated. Closed your terminal by accident? claude --resume picks up exactly where you left off. No lost context.

Hook Events

Hooks are shell commands that fire automatically during Claude Code's lifecycle. You configure them in settings.json:

Example: auto-format code after every Write:

{
  "hooks": {
    "PostToolUse": [{
      "matcher": "Write",
      "command": "prettier --write $FILE_PATH"
    }]
  }
}

Learn more in the Hooks guide.

The Commands People Sleep On

/compact is a lifesaver. Long session getting sluggish? Context window filling up? /compact compresses your conversation while preserving the important context. Use it before Claude starts forgetting things, not after. I run it every 30-40 minutes during long sessions.

/model for cost control. Start complex tasks with Opus, then switch to Haiku for simple follow-ups. "Fix the typo I just introduced" doesn't need the most expensive model. You can switch mid-conversation without losing context.

/cost for awareness. Check it every 30 minutes during long sessions. Shows per-model and cache-hit breakdown so you know exactly where your tokens are going. Most people are surprised by the breakdown.

/rewind is your undo button. Sent Claude Code in the wrong direction? /rewind takes you back one turn. Way more useful than /clear, which nukes everything. It only restores Claude's own file edits, though. Anything changed through a shell command, sed -i included, stays changed.

Escape is not Ctrl+C. Escape gracefully stops the current operation. Claude Code keeps your conversation and any partial work. Ctrl+C is the hard kill. Use it only when things are truly stuck. The difference matters.

! for quick shell commands. Need to check git status without leaving the Claude Code session? Just type ! git status. The output lands right in your conversation, and Claude Code can reference it.

"You should know" catches what you missed. It's a built-in mod, off by default. Turn it on with /plugin enable cc-plugin-you-should-know@builtin, and a side agent rereads the conversation now and then and posts one short note above the prompt, like a tradeoff Claude made without saying so. It runs on your session's model with no tools, and each check counts toward your usage. /plugin lists it only when you connect to the Claude API directly and your organization has it; set CLAUDE_CODE_YOU_SHOULD_KNOW_DEBUG=1 to watch the checks run.

Headless & CI Mode

Run Claude Code in automated pipelines:

# Pipe mode for CI/CD
claude -p "run tests and report failures" | tee test-output.txt

# Headless with specific model
claude -p --model haiku "lint all TypeScript files"

# Chain with other tools
git diff | claude -p "review this diff for bugs"

Quick Reference Card

── Session ─────────────────────────────────────
/help  /clear  /compact  /cost  /status  /usage
/quit  /export  /rewind

── Config ──────────────────────────────────────
/init  /model  /config  /permissions  /doctor
/memory  /context

── Workflow ────────────────────────────────────
/review  /plan  /agents  /mcp

── Keyboard ────────────────────────────────────
Escape (cancel)  Shift+Tab (plan)  Tab (complete)
Up (history)  Ctrl+R (search)  ! (shell)
\ (multi-line)  @ (file ref)

── CLI ─────────────────────────────────────────
-p (pipe)  -c (continue)  --resume  --verbose
--model  --effort  --bare  -r (resume)

Print this. Tape it to your monitor. You'll reference it more than you think.