agent-sandbox
Pass
Audited by Gen Agent Trust Hub on Apr 27, 2026
Risk Level: SAFE
Full Analysis
- [EXTERNAL_DOWNLOADS]: Fetches Kubernetes operator manifests and extension configurations directly from the official kubernetes-sigs GitHub repository during installation.
- [EXTERNAL_DOWNLOADS]: References the 'k8s-agent-sandbox' Python package on PyPI and the 'sigs.k8s.io/agent-sandbox' Go module for client-side integration.
- [COMMAND_EXECUTION]: Provides administrative shell commands using 'kubectl' for managing the lifecycle of sandboxed environments, including patching resource specifications and managing claim expiries.
- [COMMAND_EXECUTION]: Includes a maintenance script for 'Warm Pool Refresh' that automates the comparison and cleanup of sandbox resources to ensure they align with updated templates.
- [PROMPT_INJECTION]: The skill addresses the risk of executing untrusted agent code (Indirect Prompt Injection surface) by providing robust architectural guidance on container isolation (gVisor/Kata) and egress-controlled NetworkPolicies.
Audit Metadata