Шпаргалка Claude Code

Шпаргалка Claude Code — горячие клавиши, Slash-команды, CLI, переменные окружения, MCP, Skills, Agents

Обновлено 2026-10-08

Содержание

Команды сверяйте с локальным claude --version и официальными Releases. Сохраните эту страницу для быстрого доступа.

Keyboard Shortcuts ¶

General Controls ¶

Shortcut	Function
`Ctrl+C`	Cancel input / interrupt generation
`Ctrl+D`	Exit session
`Ctrl+L`	Clear screen
`Ctrl+O`	Toggle verbose output (view thinking)
`Ctrl+R`	Reverse search history
`Ctrl+G`	Open prompt in editor
`Ctrl+B`	Run current task in background
`Ctrl+T`	Toggle task list
`Ctrl+V`	Paste image
`Ctrl+F`	Kill background agents (press twice)
`Esc` `Esc`	Undo / go back

Mode Switching ¶

Shortcut	Function
`Shift+Tab`	Cycle permission modes (auto → Normal → Auto-accept → Plan)
`Alt+P`	Switch model
`Alt+T`	Toggle thinking mode

Input Tips ¶

Shortcut	Function
`\` + `Enter`	Newline (shortcut)
`Ctrl+J`	Newline (control sequence)

Prefix Symbols ¶

Prefix	Function
`/`	Slash commands
`!`	Execute bash directly
`@`	File mention + autocomplete

Session Selector ¶

Shortcut	Function
`↑` `↓`	Navigate
`←` `→`	Expand / Collapse
`P`	Preview
`R`	Rename
`/`	Search
`A`	All projects
`B`	Current branch

Slash commands ¶

Session Management ¶

Command	Function
`/clear`	Clear conversation
`/compact [focus]`	Compress context (optionally specify focus)
`/resume`	Resume / switch session
`/rename [name]`	Name current session
`/branch [name]`	Fork conversation (псевдоним `/fork`)
`/cost`	Token usage stats
`/context`	Visualize context usage (grid)
`/diff`	Interactive diff viewer
`/copy`	Copy last reply
`/export`	Export conversation

Configuration ¶

Command	Function
`/config`	Open settings
`/model [model]`	Switch model (`←` `→` для регулировки effort)
`/fast [on\\|off]`	Toggle fast mode
`/vim`	Toggle vim mode
`/theme`	Change color theme
`/permissions`	View / update permissions
`/effort [level]`	Set effort level（low / med / high / max / auto）
`/color [color]`	Set prompt bar color
`/keybindings`	Custom key bindings
`/terminal-setup`	Configure terminal shortcuts

Tools ¶

Command	Function
`/init`	Create CLAUDE.md
`/memory`	Edit CLAUDE.md files
`/mcp`	Manage MCP servers
`/hooks`	Manage hooks
`/skills`	List available skills
`/agents`	Manage agents
`/chrome`	Chrome integration
`/reload-plugins`	Hot-reload plugins
`/add-dir <path>`	Add working directory

Special Features ¶

Command	Function
`/btw <question>`	Side question (no context, no impact on main chat)
`/plan [desc]`	Plan mode (+ auto-start analysis)
`/loop [interval]`	Schedule recurring task (напр. `/loop 5m проверить деплой`)
`/voice`	Push-to-talk voice mode (20 languages)
`/doctor`	Diagnose installation
`/pr-comments [PR]`	Fetch GitHub PR comments
`/stats`	Usage streak and preferences
`/insights`	Session analytics report
`/desktop`	Continue to Desktop app
`/remote-control`	Connect to claude.ai/code (alias `/rc`）
`/usage`	Plan quota and rate status
`/schedule`	Cloud scheduled tasks
`/security-review`	Security analysis of changes
`/help`	Show help + all commands
`/feedback`	Submit feedback (alias `/bug`）
`/release-notes`	View full changelog

CLI Commands & Flags ¶

Core Commands ¶

Command	Function
`claude`	Interactive start
`claude "question"`	Start with initial prompt
`claude -p "query"`	Headless mode (non-interactive)
`claude -c`	Continue last conversation
`claude -r "name"`	Resume session by name
`claude update`	Update to latest version

Key Flags ¶

Flag	Function
`--model`	Set model
`-w`	Git worktree mode
`-n` / `--name`	Session name
`--add-dir`	Add extra directory
`--agent`	Use specified agent
`--allowedTools`	Pre-approve specified tools
`--output-format`	Output format（json / stream-json）
`--json-schema`	Structured JSON output
`--max-turns`	Limit interaction turns
`--max-budget-usd`	Cost cap (USD)
`--console`	Auth via Anthropic Console
`--verbose`	Verbose output
`--bare`	Minimal headless (no hooks/LSP/plugins)
`--channels`	Permission forwarding / MCP push
`--remote`	Web session
`--effort`	Set effort（low / med / high / max）
`--permission-mode`	Permission mode (default / acceptEdits / plan / auto / dontAsk / bypassPermissions)
`--dangerously-skip-permissions`	Skip all permission prompts
`--chrome`	Chrome integration

MCP Servers ¶

Add Server ¶

Flag	Transport
`--transport http`	Remote HTTP (recommended)
`--transport stdio`	Local process
`--transport sse`	Remote SSE

Scope ¶

Scope	Config File	Description
Local	`.claude.json`	Single project
Project	`.mcp.json`	Shared (commit to VCS)
User	`~/.claude.json`	Global

Management Commands ¶

Command	Function
`/mcp`	Interactive management UI
`claude mcp list`	List all servers
`claude mcp serve`	Expose Claude Code itself as MCP server

Elicitation: MCP-серверы могут запрашивать ввод пользователя в середине выполнения задачи.

Memory & File System ¶

CLAUDE.md Locations ¶

Location	Scope
`./CLAUDE.md`	Project-level (team shared)
`~/.claude/CLAUDE.md`	User-level (all projects)
`/etc/claude-code/`	Managed (organization-wide)

Rules & Imports ¶

Path / Syntax	Description
`.claude/rules/*.md`	Project rule files
`~/.claude/rules/*.md`	User rule files
`paths:` frontmatter	Path-specific rules (only for matched paths)
`@path/to/file`	Import file content in CLAUDE.md

Auto Memory ¶

~/.claude/projects/<proj>/memory/
├── MEMORY.md          ← Memory index
└── topic_xxx.md       ← Topic-based memory files

Memory files auto-loaded into context, preserved after /compact.

Workflow & Tips ¶

Plan Mode ¶

Action	Description
`Shift+Tab`	Cycle: auto → Normal → Auto-accept → Plan
`--permission-mode plan`	Start in plan mode

Thinking & Effort ¶

Action	Description
`Alt+T`	Toggle thinking mode
`"ultrathink"`	Max effort for this turn
`Ctrl+O`	Просмотр процесса размышления (Verbose output)
`/effort`	Set level: low · med · high · max

Git Worktrees ¶

Action	Description
`--worktree name`	Create isolated branch workspace per feature
`isolation: worktree`	Agent runs in isolated worktree
`sparsePaths`	Sparse checkout only needed dirs
`/batch`	Auto-create multiple worktrees for parallel changes

Voice Mode ¶

Action	Description
`/voice`	Enable push-to-talk
`Space` (удержание)	Hold to record, release to send
20 языков	англ. / исп. / фр. / нем. / кит. / яп. / кор. ...

Context Management ¶

Action	Description
`/context`	View usage + optimization tips
`/compact [focus]`	Compress with focus
Auto-compact	Auto-triggers at ~95% capacity
1M context	Opus 4.8 (доступно для Max/Team/Enterprise)
CLAUDE.md	Preserved after compact

Advanced Session Tips ¶

Action	Description
`claude -c`	Continue last conversation
`claude -r "name"`	Resume by name
`/btw question`	Side question, no main context impact

SDK / Headless Mode ¶

Action	Description
`claude -p "query"`	Non-interactive run
`--output-format json`	Structured output
`--max-budget-usd 5`	Cost cap
`cat file \\| claude -p`	Pipe input

Scheduling & Remote ¶

Action	Description
`/loop 5m msg`	Recurring task
`/rc`	Remote control
`--remote`	Web session на claude.ai

Skills & Agents ¶

Built-in Skills ¶

Skill	Function
`/simplify`	Code review (3 parallel agents)
`/batch`	Large-scale parallel changes (5-30 worktrees)
`/debug [desc]`	Debug from logs
`/loop [interval]`	Periodic scheduled tasks
`/claude-api`	Load API + SDK reference docs

Custom Skill Locations ¶

Location	Scope
`.claude/skills/<name>/`	Project skills
`~/.claude/skills/<name>/`	Personal skills

Skill Frontmatter Fields ¶

Field	Description
`description`	Auto-trigger description
`allowed-tools`	Tools that skip permission prompts
`model`	Override model for this skill
`effort`	Override effort level
`context: fork`	Run in subagent
`$ARGUMENTS`	User input placeholder
`${CLAUDE_SKILL_DIR}`	Skill directory path
!`cmd`	Dynamic context injection (run cmd, embed result)

Built-in Agents ¶

Agent	Description
Explore	Быстрый поиск только для чтения (использует Haiku)
Plan	Research for plan mode
General	Full toolset, complex tasks
Bash	Independent context terminal

Agent Frontmatter Fields ¶

Field	Description
`permissionMode`	default / acceptEdits / plan / dontAsk / bypass
`isolation: worktree`	Запуск в git worktree
`memory: user\\|project`	Persistent memory
`background: true`	Background task
`maxTurns`	Limit agentic turns
`SendMessage`	Resume stopped agents

Configuration & Environment Variables ¶

Файлы конфигурации ¶

File	Description
`~/.claude/settings.json`	User settings
`.claude/settings.json`	Project shared settings
`.claude/settings.local.json`	Local-only settings
`~/.claude.json`	OAuth, MCP, state
`.mcp.json`	Project MCP servers

Key Settings ¶

Setting	Description
`modelOverrides`	Map model selector to custom model IDs
`autoMemoryDirectory`	Custom memory directory
`worktree.sparsePaths`	Sparse checkout directories

Key Environment Variables ¶

Variable	Description
`ANTHROPIC_AUTH_TOKEN`	API key
`ANTHROPIC_MODEL`	Default model
`ANTHROPIC_BASE_URL`	API endpoint（QCode.cc: `https://api.qcode.cc/api`）
`CLAUDE_CODE_EFFORT_LEVEL`	Default effort（low / med / high）
`MAX_THINKING_TOKENS`	Thinking token limit (0 = off)
`CLAUDE_CODE_MAX_OUTPUT_TOKENS`	Output token limit (default 32K)
`ANTHROPIC_CUSTOM_MODEL_OPTION`	Custom /model dropdown entry
`CLAUDE_CODE_PLUGIN_SEED_DIR`	Plugin seed dirs (colon-separated)
`CLAUDECODE`	Detect if in Claude Code shell (=1)
`IS_DEMO`	Demo mode (hide email/org)
`CLAUDE_CODE_DISABLE_CRON`	Disable cron tasks

Permission mode ¶

Mode	Description	Flag
`auto`	Классификатор проверяет действия, требующие одобрения, и блокирует опасные (режим по умолчанию для интерактивных сессий с 2.1.283)	`--permission-mode auto`
`default`	Запрашивает подтверждение для каждого действия, требующего одобрения (в UI подписано как Manual)	`--permission-mode default`
`acceptEdits`	Auto-accept file edits	`--permission-mode acceptEdits`
`plan`	Read-only, no modifications	`--permission-mode plan`
`dontAsk`	Deny all except explicitly allowed	`--permission-mode dontAsk`
`bypassPermissions`	Skip all permission checks	`--permission-mode bypassPermissions`

CLI flag --dangerously-skip-permissions equals bypassPermissions. Use only in isolated environments. В CI и скриптах задавайте режим явно; рекомендуем --permission-mode dontAsk вместе с --allowedTools. Без него claude -p через QCode запускается в auto.

QCode.cc Quick Setup ¶

# One-line setup
export ANTHROPIC_BASE_URL=https://api.qcode.cc/api
export ANTHROPIC_AUTH_TOKEN=cr_ваш_ключ

The same cr_ key works for both Claude Code and Codex, sharing plan quota.

Связанные документы ¶

Полное руководство по Claude Code — From zero to mastery
Руководство по настройке CLAUDE.md — Project config guide
Система Hooks — Automation hooks
Автоматизация и CI/CD — Headless mode in practice
Полное руководство по Codex — Codex CLI guide
Codex vs Claude Code — Two tools compared
Тарифы и цены — QCode.cc plans

← Предыдущая

Лучшие практики безопасности

Часто задаваемые вопросы (FAQ)

Шпаргалка Claude Code

Похожие документы