skills/nvidia/skills/doca-rmax/Gen Agent Trust Hub

doca-rmax

Pass

Audited by Gen Agent Trust Hub on Oct 16, 2026

Risk Level: SAFECOMMAND_EXECUTIONPRIVILEGE_ESCALATIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to use several system tools for environment verification, device discovery, and debugging. These include pkg-config, devlink, ip, ethtool, doca_caps, dmesg, and mlxconfig. These are standard utilities for managing network drivers and hardware configurations on Linux and BlueField DPUs.
  • [PRIVILEGE_ESCALATION]: Several commands in the tasks companion are identified as requiring elevated privileges (sudo). Specifically, devlink dev show, dmesg | tail, and mlxconfig require root access to interact with the kernel and hardware registers. The instructions also mention mlnx group membership as an alternative for device access.
  • [INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it processes untrusted user input (questions about media streams) and has the capability to execute shell commands for diagnostics.
  • Ingestion points: User queries regarding stream configuration, errors (e.g., "ST 2110 receive isn't getting frames"), and jitter issues enter the agent context.
  • Boundary markers: The instructions do not define specific delimiters for interpolating user input into commands.
  • Capability inventory: The skill uses pkg-config, devlink, ip, ethtool, doca_caps, dmesg, and mlxconfig across TASKS.md.
  • Sanitization: There are no explicit instructions for sanitizing or escaping user-provided values (like device names or IP addresses) before they are used in shell commands.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 16, 2026, 05:17 AM