Claude Code Cheat Sheet

⌨️ KEYBOARD SHORTCUTS

General Controls

Ctrl C Cancel input/generation

Ctrl D Exit session

Ctrl L Repaint screen; double-press /clear removed

Ctrl O Toggle verbose transcript

Ctrl R Reverse search history

Ctrl G Open prompt in editor

Ctrl B Background running task

Ctrl T Toggle task list

CtrlX CtrlK Kill bg agents; agents view: remove session + worktree, preserve unpushed commits

CtrlAltF Toggle Focus view (VSCode: hide tool activity behind expandable summary)

Esc Esc Rewind / undo

← Open agents view

{ / } Jump between user prompts (transcript view)

? Show keyboard shortcuts (transcript view)

Mode Switching

Shift Tab Cycle permission modes

Alt P Switch model

Alt T Toggle thinking on/off

Input

\ Enter Newline (quick)

Ctrl J Newline (control seq)

v / V (vim mode) Visual / visual-line mode with selection

/ (vim NORMAL) Reverse history search (like Ctrl+R)

:shortcode: Emoji autocomplete in prompt; accepts alternate shortcodes (:thumbsup:, :love:)

Prefixes

/ Slash command

! Direct bash; file path autocomplete; Claude responds to output

@ File mention + autocomplete; @session cross-session msg (Ctrl+O expands) NEW

Session Picker

↑↓ Navigate

←→ Expand/collapse

R Rename

/ Search

CtrlA Show all projects (in /resume)

🔌 MCP SERVERS

Add Servers

--transport http Remote HTTP (recommended)

--transport stdio Local process

--transport sse Remote SSE

Scopes

Local settings.local.json (you only)

Project .mcp.json (shared/VCS)

User ~/.claude.json (global)

Manage

/mcp Interactive UI

claude mcp list List all servers; ⏸ pending approval, ⊘ Disabled (no health check)

claude mcp serve CC as MCP server

claude mcp login <name> Authenticate MCP server from CLI; --no-browser for SSH

claude mcp logout <name> Revoke MCP server authentication from CLI

Elicitation Servers request input mid-task

_meta maxResultSizeChars Override result size up to 500K

alwaysLoad Skip tool-search deferral for server tools

headersHelper Command minting HTTP headers for catalog/archive fetches; -y to auto-approve

workspace Reserved names — skipped: workspace, Claude Browser, Claude Preview

${CLAUDE_PROJECT_DIR} Reference project dir in MCP server commands

roots/list Additional working dirs advertised; notifications/roots/list_changed on changes

⚡ SLASH COMMANDS

Session

/clear Clear conversation

/compact [focus] Compact context

/resume Resume/switch session; agent view: picker of past sessions incl. deleted

/rename [name] Name current session

/branch [name] Branch conversation

/context Visualize context (grid)

/diff Interactive diff viewer (keyboard-scrollable detail)

/rewind Rewind conv / code checkpoint; resume from before /clear (/undo alias)

/recap Context summary when returning to session

/focus Toggle focus view

/goal [condition] Set completion condition; auto-clears on fatal; max 3 idle check-ins per batch NEW

/cd [path] Move session to new working dir; settings, hooks, MCP, skills apply immediately NEW

/fork Copy conversation into new background session with its own worktree

/subtask Launch in-session subagent (former /fork behavior)

Config

/config Open settings; key=value sets any setting; --help lists shorthand keys

/model [model] Switch model; Org/Role default when admin sets one (←→ effort, s = session only)

/fast [on|off] Toggle fast mode

/theme Change color theme; Auto matches terminal

/permissions View/update permissions; Auto mode tab; opens while Claude is working NEW

/effort [level] Set effort; interactive slider (low–max; Faster/Smarter labels)

/keybindings Customize keyboard shortcuts

/terminal-setup Configure terminal keybindings

/login Sign in: Console account (keyless, recommended) or API key NEW

Tools

/init Create CLAUDE.md

/memory Edit CLAUDE.md files

/mcp Manage MCP servers

/hooks Manage hooks

/skills List available skills

/reload-plugins Hot-reload plugins

/reload-skills Re-scan skill directories without restarting

/add-dir <path> Add working directory

/plugin [name] Browse, install, enable/disable plugins and marketplaces

/plugin list [--enabled|--disabled] List installed plugins with status filters

/ide Connect to IDE for diagnostics and editor integration

Special

/btw <question> Side question (no context); bare reopens panel; ←/→ browse answers

/plan [desc] Plan mode (+ auto-start)

/loop [interval] Recurring task (/proactive alias)

/bg [prompt] Fork current turn into an attachable background session

/voice Push-to-talk voice (20 langs)

/doctor Setup checkup: diagnose/fix issues; suggest CLAUDE.md trim (/checkup alias)

/remote-control Bridge to claude.ai/code (/rc); auto-start only via user-scope settings

/usage Usage stats with per-category breakdown, cost, rate status; loops breakdown NEW

/schedule Cloud scheduled tasks

/security-review Security analysis of changes

/usage-credits View usage credits (renamed from /extra-usage)

/feedback Submit feedback; include recent sessions (alias: /bug)

/workflows View dynamic workflow runs; press f to filter status

/status Session status, warnings, session kind; skipped sources; GitHub web connection NEW

/cost Show cost estimates; includes data-residency premium where applicable

/web-setup Connect GitHub for Claude Code on the web (Pro/Max) NEW

📁 MEMORY & FILES

CLAUDE.md Locations

./CLAUDE.md Project (team-shared)

~/.claude/CLAUDE.md Personal (all projects)

/etc/claude-code/ Managed (org-wide)

Rules & Import

.claude/rules/*.md Project rules

~/.claude/rules/*.md User rules

paths: frontmatter Path-specific rules

@path/to/file Import in CLAUDE.md

Auto Memory

~/.claude/projects/<proj>/memory/ Auto-loaded per project

MEMORY.md Memory index + topic files; errors over read limit (loaded content only)

modified (frontmatter) ISO timestamp in memory file frontmatter

💡 WORKFLOWS & TIPS

Plan Mode

Shift Tab Manual → Auto-Accept → Plan

⏸ footer badge Shows active manual permission mode

--permission-mode plan Start in plan mode

Auto mode Max & Bedrock/Vertex/Foundry; disableAutoMode; classifier adjudicates edge cases

Thinking & Effort

Alt T Toggle thinking on/off

"ultrathink" Max effort for turn

Ctrl O Toggle verbose transcript

/effort ○ low · ◐ med · ◑ xhigh · ● high/max (Faster↔Smarter)

Git Worktrees

--worktree name Isolated branch per feature

isolation: worktree Agent in own worktree; auto-commits, pushes, opens draft PR

sparsePaths Checkout only needed dirs

/batch Auto-creates worktrees

EnterWorktree (switch) Switch between Claude-managed worktrees; confirms if outside project dir

Voice Mode

/voice Enable push-to-talk

Space (hold) Record, release to send

20 languages EN, ES, FR, DE, CZ, PL…

Context Management

/context Usage + optimization tips; warns if exceeding context window

/compact [focus] Compress with focus

Auto-compact ~95% capacity; 1M-window → ~967K NEW

Summarize up to here Rewind menu option — compress earlier context, keep recent turns

1M context Sonnet 5 (default) + Opus 5 + Opus 4.8 (Max/Team/Ent)

CLAUDE.md Survives compaction!

Session Power Moves

claude -c Continue last conv

claude -r "name" Resume by name

/btw question Side Q, no context cost

SDK / Headless

claude -p "query" Non-interactive

--output-format json Structured output

--max-budget-usd 5 Cost cap; also stops background subagents when reached

cat file | claude -p Pipe input

Auto-continue (-p) Non-interactive sessions auto-continue responses cut off mid-stream NEW

Scheduling & Remote

/loop 5m msg Recurring task (/proactive)

/rc Remote Control

--remote Web session

Push notifications Mobile push via Remote Control

API key → no cloud API key / non-Anthropic ANTHROPIC_BASE_URL disables RC, /schedule, MCP, notifications

Dynamic workflows Orchestrates agents; default medium (<15); triggers on 'run a workflow', 'workflow:'

Self-hosted runner Own machines/containers as CC session hosts; claude self-hosted-runner (Team/Ent)

Write tool Newer models can overwrite unread files, matching Edit tool rules

🖥️ CLI & FLAGS

Core Commands

claude Interactive

claude "q" With prompt

claude -p "q" Headless

claude -c Continue last

claude -r "n" Resume

claude update Update

claude plugin init <name> Scaffold a new plugin in .claude/skills

claude agents Agent dashboard; MRs as !N; --cwd, --add-dir, --settings, --mcp-config, --model, --effort, --json

claude agents --json List live sessions as JSON; --all includes completed; id/state fields

claude --bg --exec '<cmd>' Run shell command as attachable background session

claude plugin enable <name> Enable a plugin; force-enables its dependencies

claude plugin disable <name> Disable a plugin; refuses if a dependent is enabled

claude auto-mode reset Restore default auto-mode config; --yes skips confirmation

claude self-hosted-runner CC hosts; --base-dir, --defer-shutdown-max-min, --proxy-auth-cmd/file (Team/Ent)

archive (plugin source) Install plugins from zip over HTTPS; SHA-256 pinning

claude remote-control --continue Resume most recent Remote Control session

Key Flags

--model Set model

-w Git worktree; GitLab MR URLs

-n / --name Session name

--add-dir Add dir

--agent Use agent

--allowedTools Pre-approve

--output-format json/stream

--json-schema Structured

--max-turns Limit turns

--max-budget-usd Cost cap; also stops background subagents when reached

--verbose Verbose

--remote Web session

--from-pr Load PR/MR from GitHub/GitLab/Bitbucket/GHE

--effort low/med/xhigh/high/max

--permission-mode plan/manual/…

--dangerously-skip-permissions Skip all prompts; catastrophic rm (incl. $()/backticks) still prompts ⚠️

--chrome Chrome

--fallback-model Fallback when primary unavailable (interactive + headless)

--thinking disabled turns off thinking on default-thinking models

--plugin-dir Load plugin from directory or .zip archive

--safe-mode Start with all customizations disabled for troubleshooting

--ax-screen-reader Opt-in plain-text rendering for screen reader users

--teleport <session id> Continue a cloud session locally

🤖 SKILLS & AGENTS

Built-in Skills

/verify Verification; no longer auto-invoked by Claude

/code-review [level] [pr#] Reviews diff or PR; remembers effort; --comment, --fix; high+ runs in bg; ultra=cloud

/deep-research [topic] Deep research; no longer auto-invoked by Claude

/batch Large parallel changes (5-30 worktrees)

/debug [desc] Troubleshoot from debug log

/loop [interval] Recurring task (/proactive alias)

/claude-api API + SDK reference; cost-optimize subcmd; Admin API coverage; Python 0.x→1.x NEW

/ultrareview [PR#] Cloud code review (parallel multi-agent)

/less-permission-prompts Scan transcripts for allowlist proposals

/simplify Cleanup-only review (reuse, simplify, efficiency, altitude) and apply fixes

/dataviz Chart and dashboard design guidance; runnable color-palette validator

Custom Skill Locations

.claude/skills/<name>/ Project skills (auto-loaded as plugins)

~/.claude/skills/<name>/ Personal skills

Nested .claude/skills Nested skill dirs load; clashes show as <dir>:<name>

Nested .claude/ priority Agent/workflow/output-style closest to cwd wins on collision

Stacked invocation Chain up to 5 skills: /skill-a /skill-b do XYZ

Skill Frontmatter

description Auto-invocation trigger

allowed-tools Skip permission prompts

disallowed-tools Remove tools from model while skill is active

model Override model for skill

effort Override effort level

context: fork Run in background subagent by default; background: false to opt out

$ARGUMENTS User input placeholder

${CLAUDE_SKILL_DIR} Skill's own directory

!`cmd` Dynamic context injection

bin/ Plugin ships executables

keep-coding-instructions Frontmatter for plugin output styles

monitors Plugin background monitors (auto-arm on session/skill)

Built-in Agents

Explore Fast read-only (inherits session model, capped at opus)

Plan Research for plan mode

General Full tools, complex tasks

Bash Terminal separate context

Workflow Dynamic multi-agent orchestration (opt-in); /workflows to view runs

Nesting depth Sub-agents nest up to depth 3; set CLAUDE_CODE_MAX_SUBAGENT_SPAWN_DEPTH=1 to disable

Agent Frontmatter

permissionMode default/acceptEdits/plan/dontAsk/bypass

isolation: worktree Run in git worktree

memory: user|project Persistent memory

background: true Background task; non-teammate spawns default to bg in interactive

maxTurns Limit agentic turns; partial output + SendMessage hint when reached NEW

SendMessage Resume agents; ListAgents incl. teammates; bare-name delivery; RC by name; refuses oversized/burst; notify_when_idle

initialPrompt Auto-submit first turn

subagent_type: fork Inherit full conversation + prompt cache; on by default

⚙️ CONFIG & ENV

Config Files

~/.claude/settings.json User settings

.claude/settings.json Project (shared)

.claude/settings.local.json Local only

~/.claude.json OAuth, MCP, state

.mcp.json Project MCP servers

Key Settings

modelOverrides Map model picker → custom IDs

worktree.sparsePaths Sparse checkout dirs

skillOverrides Control skill visibility: off/user-invocable-only/name-only

disableAllHooks Disable all hooks via settings.json / managed settings

fallbackModel Up to 3 fallback models tried in order when primary is overloaded

deny: tool-name glob * denies all; Tool(param:value) matches; Write/Glob/NotebookEdit→Edit/Read; dir/** cwd-only in hooks

disableBundledSkills Hide bundled skills, workflows, and built-in slash commands from model

sandbox.credentials Block credentials; mask, extract, JWT maskClaims, AWS SigV4; needs tlsTerminate

sandbox.network.strictAllowlist Deny non-allowlisted hosts for sandboxed commands without prompting

disableAutoMode Disable auto mode in settings.json

"owner/*" marketplaces Wildcards & bare URLs in strict/blocked; allowed/additional aliases; GitLab repos

Concise output style Built-in style: leads with results, skips preamble; select in /config Output style

keybindingFlavor readline = Bash word keys (Ctrl+W, Alt+F/D, Ctrl/Option+→); classic (default)

modelPicker Curate /model picker with ordered, labeled models; append or replace built-in lineup NEW

promptCacheTtl / subagentPromptCacheTtl 1-hour prompt cache on main conv; subagents at 5 min (API-key/cloud) NEW

modelPricing Org contracted per-model rates + discount for /cost, status line, telemetry NEW

feedbackDrafts SendFeedback tool drafts feedback for /feedback review; false to disable NEW

Key Env Vars

ANTHROPIC_API_KEY API key

ANTHROPIC_MODEL Forces model; /model picks don't persist; org restrictions can override

ANTHROPIC_DEFAULT_MODEL Set initial model for new sessions; /model picks persist across restarts

CLAUDE_CODE_EFFORT_LEVEL low/med/high

MAX_THINKING_TOKENS 0=off; disables thinking on models that think by default

CLAUDE_EFFORT Current effort level in hooks and Bash tool

CLAUDE_PROJECT_DIR Project dir passed to MCP stdio servers and hooks env

CLAUDE_CODE_WORKFLOWS Enable Workflow tool for multi-agent orchestration

CLAUDE_CODE_SUBAGENT_MODEL Set model for subagents and agent-team teammates

CLAUDE_CODE_MAX_CONCURRENT_SUBAGENTS Cap on concurrently-running subagents (default 20)

CLAUDE_CODE_MAX_SUBAGENT_SPAWN_DEPTH Nested subagent spawning; default depth 3; set 1 to disable

CLAUDE_CODE_DISABLE_1M_CONTEXT Hold 1M-window models to 200K via auto-compaction

Hooks

PreToolUse Before tool executes

PostToolUse After tool executes (duration_ms; can replace output)

Notification When Claude sends notification; agent_needs_input / agent_completed from bg agents

Stop When Claude finishes response (background_tasks, session_crons)

SubagentStop When subagent finishes (background_tasks, session_crons)

mcp_tool type Invoke MCP tool directly from hook

args: string[] Hook exec form — spawn directly without shell

continueOnBlock PostToolUse: feed rejection reason back, continue turn

hookSpecificOutput.additionalContext Stop/SubagentStop: give Claude feedback, continue turn

SessionStart Run on session start/resume; source="fork" for forks; set title, reload skills

ConfigChange Fire when settings files change (hot-reload)

DirectoryAdded Fires after /add-dir or SDK register_repo_root registers new working dir