Claude Code Cheat Sheet 2026 — Commands, Shortcuts, Settings

Bookmark this page. The tables below are ordered by frequency-of-use, not alphabetically, so scanning top-down covers 80% of daily work. Every entry maps to Claude Code 2026 (late Q1) behavior — if your version differs, claude --version first.

Print-friendly: the page is single-column and avoids color-dependent cues. Ctrl+P (Cmd+P on macOS) produces a clean two-page PDF.

Session Control

Command	What it does
/help	List all slash commands in current session
/clear	Reset conversation, keep env
/compact	Manually compress context to free tokens
/resume	Resume the last suspended session
/exit or /quit	Close Claude Code
/logout / /login	Switch Anthropic account

Configuration & Diagnostics

Command	What it does
/config	Open the settings UI
/model	Switch model (Opus / Sonnet / Haiku)
/doctor	Run environment diagnostics
/cost	Show current session cost
/status	Show model, version, working dir, active agents
/permissions	Inspect effective permission set
/init	Scaffold CLAUDE.md in the current repo

Advanced

Command	What it does
/hooks	List active hooks by event
/agents	List custom sub-agents
/mcp	Manage MCP servers (list, enable, disable)
/memory	Open memory index
/pr_comments	Fetch review comments on the current PR
/ide	Connect / disconnect IDE integration

Custom commands live in .claude/commands/<name>.md and appear in the same autocomplete. See the .claude/ folder guide for the template.

Key	Action
Enter	Submit prompt
Shift + Enter / Option + Enter	Insert newline (multi-line prompt)
Esc	Cancel the current tool call / interrupt stream
Double Esc	Exit the entire session
Ctrl + C	Copy selection (or cancel, second press)
Ctrl + L	Clear the terminal screen (keeps context)
Ctrl + R	Reverse-search prompt history
↑ / ↓	Navigate prompt history
Tab	Autocomplete file path / slash command
Ctrl + J	Insert newline (alt binding on some terminals)

Prefix Shortcuts in the Prompt

Prefix	Effect
@filename	Attach a file to the prompt by path
/command	Invoke a slash command
#memory note	Append a note to the memory index without running a tool

Common Invocation Flags

Flag	Purpose
--model <id>	Start in a specific model (e.g. claude-sonnet-4-6)
--verbose	Print resolution chain for permissions / hooks / memory
--print "<prompt>"	One-shot mode: execute the prompt and exit
--resume	Reopen the last session
--dangerously-skip-permissions	Disable all permission prompts — use only in sandboxes
--verbose --debug	Full diagnostic output for bug reports

Environment Variables

Variable	Effect
ANTHROPIC_API_KEY	Override OAuth with a raw API key
CLAUDE_PROJECT_DIR	(Set by Claude inside hooks) repo root
CLAUDE_SESSION_ID	(Set by Claude inside hooks) current session ID
NO_COLOR=1	Disable ANSI color output

The four fields you will touch 95% of the time:

permissions

"permissions": {
  "allow": ["Bash(npm run *)", "Edit", "Write", "Read", "Grep"],
  "deny":  ["Bash(rm -rf *)", "Bash(git push --force*)", "Bash(sudo *)"]
}COPY

hooks

"hooks": {
  "PreToolUse": [
    {
      "matcher": "Edit|Write",
      "hooks": [{ "type": "command", "command": "sh .claude/hooks/check.sh" }]
    }
  ]
}COPY

Events: PreToolUse, PostToolUse, UserPromptSubmit, Stop, SubagentStop, Notification, PreCompact, SessionStart. See the hooks cookbook for patterns.

mcpServers

"mcpServers": {
  "github": {
    "command": "npx",
    "args": ["-y", "@modelcontextprotocol/server-github"],
    "env": { "GITHUB_PERSONAL_ACCESS_TOKEN": "$GITHUB_TOKEN" }
  }
}COPY

env (per-project)

"env": {
  "PYTHONPATH": "./src",
  "NODE_ENV": "development"
}COPY

Resolution order: ~/.claude/settings.json → .claude/settings.json → .claude/settings.local.json. Later wins on key collision.

Goal	Prompt Pattern
Code review a branch	/review (if configured) or "Compare current branch to main. Report security, perf, test coverage issues as numbered list with file:line"
Debug a failing test	"Run npm test -- <file>, identify root cause, propose minimal fix — do not apply yet"
Refactor safely	"Plan the refactor first. Do not touch code. Once I approve, execute step by step."
Write a commit message	"Stage the current diff and draft a conventional commit message — do not run git commit"
Explore unfamiliar code	@src/module/ "Give me a one-paragraph architecture overview and the three entry points"
Parallel tasks	"Delegate type-checking to one agent and linting to another, then merge results"
Session handoff	/compact then "Summarize what we decided this session for next time"

Escape Hatches

• Claude loops on the same fix: Esc, then /clear and paste the last known good state.
• Context too full: /compact to squash, or /clear to reset.
• Permission prompts every 10 seconds: edit .claude/settings.json to allowlist the pattern.

This sheet is the index. Go deeper here:

• .claude/ folder complete guide — settings.json, commands, agents, hooks in depth.
• Hooks cookbook — 5 copy-paste hook recipes.
• Top 10 MCP servers — which ones to install first.
• Hidden features — KAIROS, ULTRAPLAN, Undercover Mode.

Save, print, pin to your terminal. Tool home for UUID lookup and species guide.

Can I print this cheat sheet as PDF?

Yes — Cmd + P (macOS) or Ctrl + P (Windows/Linux) opens your browser's print dialog. Choose "Save as PDF." The layout is single-column and print-friendly, producing roughly two US Letter pages.

Are these commands the same on Windows PowerShell?

Slash commands and keyboard shortcuts are identical. CLI flags on Windows use the same syntax if you launch Claude Code from WSL or an MSYS/Git Bash terminal. Native PowerShell may require escaping quotes differently — prefer WSL for consistency.

How do I add a slash command to this list?

Create .claude/commands/<name>.md with frontmatter describing it. It appears in /help and tab-autocomplete alongside the built-ins. The .claude/ guide has the exact frontmatter schema.

Why does /cost show a different number than the Anthropic console?

The in-session counter is local and tracks current session only. The console aggregates across sessions with a short delay. For exact billing, trust the console. For per-session awareness, trust /cost.

What's the difference between /clear and /compact?

/clear wipes the conversation entirely — Claude starts from scratch with your CLAUDE.md but no context. /compact preserves conclusions but discards verbose intermediate exchanges. Use /compact when you want continuity, /clear when you want a clean slate.