ascend-avi-vnpu
Pass
Audited by Gen Agent Trust Hub on Oct 10, 2026
Risk Level: SAFECOMMAND_EXECUTIONPRIVILEGE_ESCALATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides instructions for the agent to execute powerful system commands including the npu-smi utility for hardware management and the kill command for process termination. \n
- Evidence: SKILL.md and references/security-checklist.md provide command templates for npu-smi and kill -9. \n- [PRIVILEGE_ESCALATION]: The skill guides the agent through operations that typically require root or administrative privileges, such as modifying NPU virtualization modes and terminating system processes. \n
- Evidence: references/security-checklist.md explicitly notes that killing processes requires root permissions. \n- [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes output from local system tools, creating a potential attack surface if those outputs can be manipulated by an attacker. \n
- Ingestion points: npu-smi info -t template-info, npu-smi info -t proc-mem (SKILL.md). \n
- Boundary markers: The skill includes some instructions to verify template names if they are ambiguous, but lacks formal delimiters for tool output. \n
- Capability inventory: System configuration (npu-smi set) and process termination (kill -9). \n
- Sanitization: No explicit sanitization or filtering of tool-provided strings is documented before the agent uses them for decision-making.
Audit Metadata