provision-agent-workspace
Pass
Audited by Gen Agent Trust Hub on Sep 26, 2026
Risk Level: SAFENO_CODEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill consists purely of markdown documentation providing guidelines for infrastructure isolation and workspace provisioning. No executable scripts, binary files, or network-active components are included in the skill content.
- [NO_CODE]: The skill does not provide any functional code or configuration that would execute on the host or in a provisioned environment; it acts solely as an instructional reference for the agent to follow using external tools.
- [INDIRECT_PROMPT_INJECTION]: The skill defines procedures for an agent to manage environments that may execute untrusted code, representing a potential attack surface. However, the instructions prioritize security by mandating isolation boundaries, resource limits, and strict network policies as primary controls.
- Ingestion points: User requests for sandbox or VM provisioning (SKILL.md).
- Boundary markers: Explicit requirements for isolation, identity boundaries, and network policy restrictions (references/acceptance.md, references/provider-selection.md).
- Capability inventory: Basic filesystem operations and command execution are performed inside the provisioned environment for verification purposes (references/recipes.md).
- Sanitization: Focused on architectural isolation (sandboxing/microVMs) to contain the impact of processed data.
Audit Metadata