CLI Nodes¶

CLI nodes enable secure execution of command-line tools within workflows. The primary use case is running CLI-based AI agents like OpenAI Codex CLI, Claude Code, Aider, and similar tools in a sandboxed environment.

Overview¶

Modern AI coding assistants increasingly ship as CLI tools that can: - Analyze and modify codebases - Generate code from natural language - Debug and fix issues - Run automated refactoring

m9m's CLI Execute node allows you to orchestrate these tools within workflows while maintaining security through Linux sandboxing.

CLI Execute Node¶

Execute any CLI command with configurable sandboxing, isolation, and resource limits.

Type¶

n8n-nodes-base.cliExecute

Parameters¶

Parameter	Type	Required	Default	Description
`command`	string	Yes	-	The command to execute
`args`	array	No	`[]`	Command arguments
`env`	object	No	`{}`	Environment variables (key-value pairs)
`workDir`	string	No	-	Working directory for execution
`shell`	boolean	No	`false`	Run command through shell
`sandboxEnabled`	boolean	No	`true`	Enable sandbox isolation
`isolationLevel`	string	No	`standard`	Isolation level
`networkAccess`	string	No	`host`	Network access mode
`timeout`	number	No	`60`	Timeout in seconds
`maxMemoryMB`	number	No	`512`	Max memory in MB
`inputFromPrevious`	boolean	No	`false`	Send input data to stdin
`outputFormat`	string	No	`text`	Output format
`additionalMounts`	array	No	`[]`	Additional filesystem mounts

Running CLI AI Agents¶

Supported Agents¶

The CLI Execute node works with any CLI-based AI agent:

Agent	Command	Description
Claude Code	`claude`	Anthropic's CLI coding assistant
OpenAI Codex CLI	`codex`	OpenAI's code generation CLI
Aider	`aider`	AI pair programming in terminal
GitHub Copilot CLI	`gh copilot`	GitHub's CLI assistant
Continue	`continue`	Open-source AI code assistant
Cursor CLI	`cursor`	Cursor editor's CLI mode

Example - Claude Code¶

Run Claude Code to analyze a codebase:

{
  "name": "Analyze with Claude Code",
  "type": "n8n-nodes-base.cliExecute",
  "parameters": {
    "command": "claude",
    "args": [
      "--print",
      "--prompt", "Analyze this codebase and identify potential security vulnerabilities"
    ],
    "sandboxEnabled": true,
    "isolationLevel": "strict",
    "networkAccess": "host",
    "timeout": 300,
    "maxMemoryMB": 2048,
    "additionalMounts": [
      {
        "source": "/home/user/myproject",
        "destination": "/workspace",
        "readWrite": false
      }
    ],
    "env": {
      "ANTHROPIC_API_KEY": "={{ $credentials.anthropicApi.apiKey }}"
    }
  }
}

Example - OpenAI Codex CLI¶

Generate code with Codex:

{
  "name": "Generate with Codex",
  "type": "n8n-nodes-base.cliExecute",
  "parameters": {
    "command": "codex",
    "args": [
      "--quiet",
      "--approval-mode", "full-auto",
      "Add input validation to all API endpoints"
    ],
    "sandboxEnabled": true,
    "isolationLevel": "standard",
    "networkAccess": "host",
    "timeout": 600,
    "maxMemoryMB": 2048,
    "additionalMounts": [
      {
        "source": "/home/user/project",
        "destination": "/project",
        "readWrite": true
      }
    ],
    "workDir": "/project",
    "env": {
      "OPENAI_API_KEY": "={{ $credentials.openAiApi.apiKey }}"
    }
  }
}

Example - Aider for Code Review¶

Use Aider to review and suggest improvements:

{
  "name": "Code Review with Aider",
  "type": "n8n-nodes-base.cliExecute",
  "parameters": {
    "command": "aider",
    "args": [
      "--no-git",
      "--yes",
      "--message", "Review this code for best practices and suggest improvements"
    ],
    "sandboxEnabled": true,
    "isolationLevel": "standard",
    "networkAccess": "host",
    "timeout": 300,
    "additionalMounts": [
      {
        "source": "={{ $json.repoPath }}",
        "destination": "/repo",
        "readWrite": false
      }
    ],
    "workDir": "/repo"
  }
}

Example - Multi-Agent Workflow¶

Chain multiple AI agents for complex tasks:

{
  "nodes": [
    {
      "name": "Analyze with Claude",
      "type": "n8n-nodes-base.cliExecute",
      "parameters": {
        "command": "claude",
        "args": ["--print", "--prompt", "List all functions that need tests"],
        "outputFormat": "json",
        "additionalMounts": [{"source": "/project", "destination": "/workspace", "readWrite": false}]
      }
    },
    {
      "name": "Generate Tests with Codex",
      "type": "n8n-nodes-base.cliExecute",
      "parameters": {
        "command": "codex",
        "args": ["--quiet", "Generate unit tests for: {{ $json.stdout.functions }}"],
        "additionalMounts": [{"source": "/project", "destination": "/workspace", "readWrite": true}]
      }
    }
  ]
}

Configuration Options¶

Isolation Levels¶

Level	Description	Use Case
`none`	No isolation	Development/debugging only
`minimal`	Basic filesystem isolation	Trusted tools
`standard`	+ PID namespace, resource limits	Recommended for most agents
`strict`	+ Network isolation, seccomp filtering	Untrusted code analysis
`paranoid`	+ No host filesystem access	Maximum security

Network Access Modes¶

Mode	Description	When to Use
`host`	Full network access	AI agents that need API access
`isolated`	No network	Offline analysis tools
`loopback`	Localhost only	Local services only

Output Formats¶

Format	Description
`text`	Raw text output (default)
`json`	Parse stdout as JSON object
`lines`	Split stdout into array of lines

Additional Mounts¶

Mount host directories into the sandbox:

{
  "additionalMounts": [
    {
      "source": "/home/user/project",
      "destination": "/workspace",
      "readWrite": true
    },
    {
      "source": "/home/user/.config/claude",
      "destination": "/home/sandbox/.config/claude",
      "readWrite": false
    }
  ]
}

Output Structure¶

Every execution returns:

{
  "json": {
    "exitCode": 0,
    "stdout": "...",
    "stderr": "...",
    "killed": false,
    "killReason": "",
    "duration": 1500,
    "cpuTime": 1200,
    "maxMemory": 104857600
  }
}

Field	Type	Description
`exitCode`	number	Process exit code (0 = success)
`stdout`	string/object/array	Command output (format depends on outputFormat)
`stderr`	string	Error output
`killed`	boolean	Whether process was terminated
`killReason`	string	Reason if killed (timeout, signal, etc.)
`duration`	number	Wall-clock time in milliseconds
`cpuTime`	number	CPU time used in milliseconds
`maxMemory`	number	Peak memory usage in bytes

Security¶

Sandbox Technology¶

On Linux, m9m uses bubblewrap (bwrap) for sandboxing:

Namespace isolation: PID, network, mount, user namespaces
Resource limits: cgroups v2 for memory, CPU, process count
Syscall filtering: Seccomp BPF for blocking dangerous operations
Filesystem isolation: Controlled mounts with read-only defaults

Prerequisites¶

# Ubuntu/Debian
sudo apt install bubblewrap

# Fedora/RHEL
sudo dnf install bubblewrap

# Arch Linux
sudo pacman -S bubblewrap

Default Security Settings¶

Setting	Default
Memory limit	512 MB
Timeout	60 seconds
Max processes	50
Network	Host (configurable)
Filesystem	Read-only mounts

Blocked Syscalls (strict/paranoid)¶

reboot, kexec_load - System control
mount, umount - Filesystem manipulation
pivot_root, chroot - Container escape
init_module, delete_module - Kernel modules
ptrace - Process debugging
setns - Namespace escape

Best Practices¶

For AI Agents¶

Use appropriate isolation: standard for most agents, strict for untrusted code
Set reasonable timeouts: AI agents may take minutes; use 300-600 seconds
Allocate sufficient memory: 1-2 GB for most AI agents
Enable network access: Most agents need API access to work
Mount code read-only when possible to prevent accidental modifications

For Security¶

Never disable sandbox in production
Use read-only mounts unless writes are required
Limit network access when API calls aren't needed
Set memory limits to prevent runaway processes
Review agent output before applying changes to codebase

For Reliability¶

Handle non-zero exit codes in your workflow
Parse stderr for error messages
Check killed flag to detect timeouts
Use JSON output format for structured data

Troubleshooting¶

Agent not found¶

{"exitCode": 127, "stderr": "command not found: claude"}

Solution: Ensure the agent CLI is installed and in PATH, or use absolute path.

Permission denied¶

{"exitCode": 1, "stderr": "Permission denied"}

Solution: Check mount permissions or use readWrite: true for mounts that need write access.

Timeout exceeded¶

{"killed": true, "killReason": "timeout"}

Solution: Increase timeout parameter for long-running operations.

Out of memory¶

{"killed": true, "killReason": "signal 9"}

Solution: Increase maxMemoryMB parameter.

Quick Reference¶

Parameter	Recommended for AI Agents
`isolationLevel`	`standard` or `strict`
`networkAccess`	`host` (for API access)
`timeout`	`300` - `600` seconds
`maxMemoryMB`	`1024` - `2048` MB
`outputFormat`	`json` (for structured output)