Risk boundary
Sandboxing limits what a command can see or change even when the agent is allowed to request execution.
Enterprise
Execution Sandbox
The execution sandbox constrains command, attachment, and connection-backed execution so agents can gather evidence without turning every tool call into unconstrained host access. Stricter sandboxing is most important for production commands, attachment parsing, and customer-owned environments.
What this page covers
Use stricter sandboxing when commands touch production, secrets, customer data, or untrusted attachments.
Combine sandbox policy with tool approvals, connection scopes, and command-family limits.
Keep approval prompts focused on safe metadata and intent instead of raw secrets or oversized evidence.
Operational detail
How OpsRabbit supports Execution Sandbox
The execution sandbox constrains command, attachment, and connection-backed execution so agents can gather evidence without turning every tool call into unconstrained host access. Stricter sandboxing is most important for production commands, attachment parsing, and customer-owned environments.
Approval flow
Risky actions can require live approval or saved always-allowed rules after normal access checks pass.
Attachment safety
PDF and file extraction can run inside the configured sandbox, with clear messages for scanned, encrypted, or unreadable files.
Related pages