Skip to main content

AI Coding Agents

Claude Code

Run Claude Code with access limited to your project:
Allow Claude to read your global config:
Start an interactive shell inside the sandbox:

OpenClaw

Run OpenClaw gateway with nono sandbox:
Or manually specify permissions:

Generic AI Agent

Checking Path Access

Why is a path blocked?

Check with capability context

Query from inside a sandbox

Check network access

Build Tools

Cargo (Rust)

npm/Node.js

Make

Network Operations

curl/wget

Git Operations

Multi-Directory Access

Separate Source and Output

Multiple Projects

Shared Dependencies

Debugging and Testing

Dry Run

Preview what access would be granted:

Verbose Output

Testing Sandbox Enforcement

Shell Scripts

Running a Script

Inline Commands

Configuration Files

Read-Only Config

Multiple Config Files

Using Profiles

Agent Profiles

Profile with Extra Permissions

Profile with Custom Workdir

Restrict Profile to Specific Domains

Real-World Scenarios

Code Review Agent

An agent that reads code and writes review comments:

Documentation Generator

An agent that reads source and generates docs:

Data Processing Pipeline

Cloud Agent with Credential Access

By default, ~/.aws and ~/.config/gcloud are blocked by the deny_credentials group. Use --bypass-protection to grant targeted access:

Offline Build Environment

IPC Between Sandboxed Processes

Run an MCP server and a client in separate sandboxes, communicating over localhost: