hypervisor-introspection
Pass
Audited by Gen Agent Trust Hub on Nov 14, 2026
Risk Level: SAFEPRIVILEGE_ESCALATIONPERSISTENCEDYNAMIC_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [PRIVILEGE_ESCALATION]: The skill provides detailed instructions for virtualization escapes (guest-to-host) and achieving kernel-level execution on hypervisor hosts, such as ESXi, which are high-privilege security boundary violations.
- [PERSISTENCE]: Documentation includes methods for maintaining access across hypervisor reboots, including the modification of ESXi boot scripts (
/etc/rc.local.d/local.sh) and the abuse of vSphere Auto Deploy to push persistent image profiles. - [DYNAMIC_EXECUTION]: The playbook instructs the agent to compile complex software from source (e.g., QEMU with specific CVE versions) and load custom kernel-mode research drivers (e.g., HyperPlatform) at runtime.
- [COMMAND_EXECUTION]: The skill contains a vast library of powerful administrative commands (e.g.,
esxcli,virsh,xl,qm,pvesh) used to manage virtual machines, datastores, and hypervisor configurations. - [EXTERNAL_DOWNLOADS]: Several instructions involve fetching source code and tools from external Git repositories, including GitLab (QEMU) and GitHub (LibVMI, DRAKVUF, HyperPlatform, Bitdefender KVMi).
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external hypervisor APIs, memory dumps, and configuration files, which could contain malicious payloads targeting the agent's interpretation logic.
- Ingestion points:
payloads.md,test-cases.md(vCenter/ESXi API responses, guest memory dumps, XenStore data). - Boundary markers: Absent; no explicit instructions are provided to the agent to treat external hypervisor output as untrusted or to ignore embedded instructions.
- Capability inventory:
SKILL.md(Bash, Read, Write, Edit, WebFetch). - Sanitization: Absent; the skill relies on the agent to process raw API and memory data without specific filters.
Audit Metadata