# Claude Code Quick Reference

*Shortcuts, slash commands, memory, MCP, hooks, agents, automation*

> Source: Anthropic Claude Code Documentation (docs.anthropic.com) · MIT

## Slash Commands

### Daily Commands

| Command | Description |
|---------|-------------|
| `/compact [focus]` | Compress context; keep key memory |
| `/resume` | Resume or switch sessions |
| `/context` | Visualize context usage |
| `/memory` | Open or edit memory files |
| `/permissions` | View or change approval mode |
| `/model [model]` | Switch model |
| `/effort [level]` | Set reasoning effort |

### Project & Workflow

| Command | Description |
|---------|-------------|
| `/init` | Create a project CLAUDE.md |
| `/mcp` | Manage MCP servers |
| `/hooks` | Manage hooks |
| `/agents` | Manage agents |
| `/plan [desc]` | Enter plan mode |
| `/branch [name]` | Fork conversation / branch context |
| `/config` | Open settings |

## Memory & Project Files

### Where Context Lives

| Choice | Use when | Watch for |
|--------|----------|-----------|
| `./CLAUDE.md` | You want shared repo instructions, build commands, and project conventions | Keep it durable and concise, not a dump of session chatter |
| `~/.claude/CLAUDE.md` | You want personal defaults across many projects | Avoid repo-specific rules that teammates should see |
| `/etc/claude-code/CLAUDE.md` | Your org needs managed instructions users should not bypass | Too heavy for personal workflow notes |
| `auto memory` | Claude should remember corrections and local facts over time | Machine-local and less portable than authored guidance |

### Rules & Imports

| Command | Description |
|---------|-------------|
| `.claude/rules/*.md` | Project rules file set for more focused guidance |
| `@path/to/file` | Import files inside CLAUDE.md |
| `paths:` | Path-scoped rules in frontmatter |

### CLAUDE.md vs Auto Memory

| Choice | Use when | Watch for |
|--------|----------|-----------|
| `CLAUDE.md` | You want durable instructions you author deliberately | Best for commands, repo structure, coding rules, and safety policy |
| `auto memory` | Claude should retain corrections and learned facts across sessions | Useful, but less predictable and not as portable |

## Permissions, Hooks & MCP

### Permission Modes

| Choice | Use when | Watch for |
|--------|----------|-----------|
| `plan` | You want research, planning, or architectural thinking without edits | Safest mode, but intentionally slower for execution |
| `default` | You want the normal interactive baseline | Expect prompts before riskier actions |
| `acceptEdits` | You want faster editing while still gating risky tools | Good speed trade-off, but still not full automation |
| `bypassPermissions` | You are in tightly trusted automation or fully controlled contexts | High risk; not a normal day-to-day setting |

### settings.json & Hooks

```
{
  "permissions": { "allow": ["Read", "Glob"], "deny": ["Bash(rm *)"] },
  "hooks": {
    "PreToolUse": [{
      "matcher": "Bash",
      "hooks": [{ "type": "command", "command": "check-command.sh $INPUT" }]
    }]
  }
}
```

### Hook Events & Results

| Command | Description |
|---------|-------------|
| `PreToolUse` | Before tool execution |
| `PostToolUse` | After tool execution |
| `Stop` | At response completion |
| `exit 0` | Allow tool to proceed |
| `exit 2` | Block tool execution |
| `stdout JSON` | Structured feedback back to Claude |

### MCP Commands

```
claude mcp add my-server -- npx my-mcp-server
claude mcp add --transport http sentry https://mcp.sentry.dev/mcp
claude mcp list
claude mcp get my-server
claude mcp remove my-server
claude mcp serve
```

### MCP Scope

| Choice | Use when | Watch for |
|--------|----------|-----------|
| `project` | The repo should share the same MCP server config | Put it in `.mcp.json` so the setup travels with the project |
| `user` | You want personal MCP servers across many repos | Lives in `~/.claude.json`; teammates will not inherit it |
| `stdio` | The server runs as a local process on your machine | Best for local tooling; install/runtime is your responsibility |
| `http / sse` | You connect to remote MCP servers over the network | Credentials, latency, and remote availability matter more |

## CLI & Automation

### Core Commands

```
claude                  # interactive
claude "query"          # start with prompt
claude -p "query"       # headless
claude -c               # continue last
claude -r "name"        # resume named session
```

### Key Flags

| Command | Description |
|---------|-------------|
| `-p` | Headless / print mode |
| `--output-format` | text / json / stream / stream-json |
| `--json-schema` | Constrain structured output |
| `--model` | Set model ID or shortcut |
| `--effort` | low / medium / high / max / auto |
| `--allowedTools` | Pre-approve restricted tool set |
| `--max-turns` | Cap turns in automation |

### Automation Examples

```
claude -p "summarize main.py" --output-format json
claude -p "list TODOs" --json-schema todo.schema.json
git diff | claude -p "review this diff"
cat log.txt | claude -p "explain the failure"
```

## Agents, Skills & Worktrees

### Choose The Execution Path

| Choice | Use when | Watch for |
|--------|----------|-----------|
| `/btw` | You want a side question using the current conversation context | No tools; it cannot inspect the repo or gather new facts |
| `Explore agent` | You want fast read-only investigation | Good for research, not normal editing |
| `General agent` | You want a delegated worker with normal tool access | Fresh context means you must specify the task clearly |
| `Skill` | You want a reusable workflow or parameterized prompt bundle | Better for repeated tasks than ad hoc questions |
| `/batch` | You want parallel worktree execution across subtasks | Higher overhead; best when the task splits cleanly |

### Worktrees & Skill Fields

| Command | Description |
|---------|-------------|
| `--worktree name` | Run session in dedicated git worktree |
| `isolation: worktree` | Agent frontmatter for worktree isolation |
| `background: true` | Run agent in background |
| `maxTurns` | Limit agentic turn count |
| `allowed-tools` | Pre-approve tools inside a skill |
| `context: fork` | Run the skill in a subagent |
| `$ARGUMENTS` | User input placeholder in a skill |

## Keyboard & Input

### High-Value Keys

| Command | Description |
|---------|-------------|
| `Ctrl+C` | Cancel current input or generation |
| `Ctrl+G` | Open the current prompt in your editor |
| `Ctrl+T` | Toggle the task list |
| `Shift+Tab` | Cycle permission modes |
| `\ + Enter` | Insert a quick newline |
| `/ / ! / @` | Command, shell, and file-path prefixes |

## Config & Defaults

### Config Files

| Choice | Use when | Watch for |
|--------|----------|-----------|
| `.claude/settings.json` | The project should share the same defaults | Commit only when the team should inherit the behavior |
| `.claude/settings.local.json` | You want local-only project overrides | Keep personal tweaks and secrets out of shared config |

### Context Moves

| Command | Description |
|---------|-------------|
| `/compact` | Compress context near capacity |
| `/context` | See token or context usage |