> ## Documentation Index
> Fetch the complete documentation index at: https://mintlify.com/eugene1g/agent-safehouse/llms.txt
> Use this file to discover all available pages before exploring further.

# Claude Code

> Anthropic's proprietary CLI/TUI agent with comprehensive sandboxing

## Overview

Claude Code is Anthropic's closed-source AI coding agent with CLI/TUI, VS Code extension, and desktop app modes. Notable for having built-in sandbox support.

<Info>
  **Binary path:** `~/.local/bin/claude`\
  **Config:** `~/.claude/`\
  **Dependencies:** `$$require=55-integrations-optional/keychain.sb,browser-native-messaging.sb$$`
</Info>

## Sandbox Profile

```scheme theme={null}
(allow file-read* file-write*
    (home-prefix "/.local/bin/claude")
    (home-subpath "/.cache/claude")
    (home-subpath "/.claude")
    (home-prefix "/.claude.json")
    (home-literal "/.claude.lock")
    (home-subpath "/.config/claude")
    (home-subpath "/.local/state/claude")
    (home-subpath "/.local/share/claude")
    (home-literal "/.mcp.json")
)

(allow file-read*
    (home-prefix "/.claude.json.")
    (home-literal "/Library/Application Support/Claude/claude_desktop_config.json")
    (subpath "/Library/Application Support/ClaudeCode/.claude")
    (literal "/Library/Application Support/ClaudeCode/managed-settings.json")
)
```

## Running in Safehouse

```bash theme={null}
# With keychain and browser integration
safehouse --enable=keychain,browser-native-messaging,network --add-dirs="$PWD" -- claude

# Chrome integration for web workflows
claude --chrome  # requires browser-native-messaging
```

## Built-in Sandbox (v2.0.24+)

Claude Code has its own Bash sandbox (Linux & macOS):

```json theme={null}
{
  "sandbox": {
    "enabled": true,
    "autoAllowBashIfSandboxed": false,
    "network": {
      "allowedDomains": [],
      "allowLocalBinding": false
    }
  }
}
```

<Note>
  Claude Code's internal sandbox applies **only to Bash tool commands**, not to Read, Write, Edit, MCP tools, or hooks.
</Note>

## Plugin System

Claude Code has a rich plugin architecture:

* **Commands** -- Custom slash commands (`.md` files)
* **Agents** -- Custom subagent definitions
* **Skills** -- Auto-activating capabilities (`SKILL.md`)
* **Hooks** -- Lifecycle event scripts
* **MCP servers** -- External tool integrations

## Enterprise Features

* MDM managed preferences (`com.openai.codex` on macOS)
* Managed settings for org-level policies
* Bash sandbox with network allowlisting
* Hook and plugin restrictions