Skip to main content
This guide explains how to run the iCommerce agent inside StateSet Sandbox for secure, isolated execution.

​
Why Sandbox

  • Isolated runtime for agent execution
  • Controlled access to tools and data
  • Deterministic workflows with auditability

​
High-Level Flow

  1. Provision a sandbox environment.
  2. Start the iCommerce agent process.
  3. Use MCP tools or the CLI for queries and actions.
  4. Stream outputs and capture logs.

​
Sandbox Setup

Create a sandbox and run a simple command:

​
Run the Agent

Inside the sandbox, run the stateset CLI in read-only mode first, then apply actions explicitly:

​
Logging and Auditing

  • Stream command output for review
  • Record actions, inputs, and outcomes
  • Keep explicit write intent for audit trails
The sandbox is an isolation boundary for the agent’s execution, not for your data: the agent still holds real credentials and acts on real commerce records. Sandboxing means a bad command cannot reach the rest of your infrastructure — it does not mean a wrong refund is reversible. Keep --apply off until a read-only pass looks right.
Last modified on August 31, 2026