nemoclaw-user-configure-security

To evaluate sandbox security controls, map each configurable knob to its default posture and the concrete risk of relaxing it, ensuring safe trade-offs when hardening network or filesystem access. This framework clarifies what each control protects and what breaks when modified.

OpenClaw protects beyond NemoClaw by independently handling prompt injection defenses, tool access control, rate limiting, and security audit coverage. Provider credentials are stored in a single system of record in OpenShell, separate from the sandbox posture decisions.

To decide on allowing additional egress for a sandbox integration, review the risk framework mapping each control to its default posture and the concrete runtime risk of relaxing it. This helps determine safe trade-offs when debugging sandbox behavior or adding integrations.

Relaxing deny-by-default process and inference posture controls risks breaking layered security protections across network, filesystem, process, and inference boundaries. Operators must understand what each control protects and what breaks at runtime before approving modifications.

Yes, operator approval is required to modify NemoClaw's default security posture because the framework uses deny-by-default behavior across network, filesystem, process, and inference layers. This ensures operators explicitly authorize any changes to sandbox controls.

The credential storage system uses OpenShell as a single system of record for provider credentials, while NemoClaw sandbox controls manage the operational security posture. This separation ensures credential protection is handled independently from configurable sandbox settings.