Skip to main content

​
Assets to protect

  • The host and runtime environment.
  • Credentials and identity material.
  • Private source code and data.
  • External systems reachable from the sandbox.
  • Other sandboxes and tenants.

​
Primary threat

Agent-generated code is untrusted and may be incorrect, overly broad or deliberately manipulated by external input. The runtime must assume that a workload may attempt to inspect its environment, consume excessive resources or reach unintended destinations.

​
Security objectives

  • Keep the workload away from the developer host.
  • Limit the workload to its assigned execution environment.
  • Expose only the resources needed for the task.
  • Destroy temporary environments reliably.
  • Record enough execution state to investigate failures and abuse.

​
Current state

jhansi.io currently uses Docker containers as the isolation backend. This provides process, filesystem and network separation at the container level. It does not provide microVM or hardware virtualisation boundaries. A production security claim requires implementation details, hardening guidance and independent testing.