The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes the vm_guest_exec and vm_guest_exec_output tools, which allow the agent to execute arbitrary shell commands within guest virtual machines. This is a high-privilege capability required for guest operations but involves significant risk if misused.
[EXTERNAL_DOWNLOADS]: The deploy_vm_from_ova and deploy_vm_from_template tools facilitate the deployment of virtual appliances from local or network paths. This involves introducing external software packages into the virtual infrastructure.
[DATA_EXFILTRATION]: The skill provides vm_guest_download and vm_guest_upload tools that allow the movement of files between the host environment and guest virtual machines, creating a potential path for data extraction from isolated systems.
[PROMPT_INJECTION]: The skill processes untrusted data from external sources including vSphere events and ESXi host logs. This creates a surface for indirect prompt injection attacks. The skill implements specific mitigations for this:
Ingestion points: Data enters the agent context through tools that query vCenter alarms, events, and host logs.
Boundary markers: External content is wrapped in explicit markers such as [VSPHERE_EVENT] and [VSPHERE_HOST_LOG] to help the agent distinguish system data from instructions.
Capability inventory: The skill possesses high-impact capabilities including VM deletion (vm_delete), power state changes, and command execution (vm_guest_exec).
Sanitization: Instructions specify that vSphere-sourced content is truncated and stripped of control characters before being presented to the agent.

vmware-aiops