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

doca-dpa

Warn

Audited by Gen Agent Trust Hub on Oct 15, 2026

Risk Level: MEDIUMPRIVILEGE_ESCALATIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [PRIVILEGE_ESCALATION]: The skill instructs the use of sudo dmesg in the TASKS.md Command Appendix to troubleshoot driver-level issues, requiring elevated permissions.
  • [COMMAND_EXECUTION]: The skill utilizes shell commands such as pkg-config, dpacc, and doca_caps for environment inspection and application building throughout the TASKS.md workflow.
  • [DYNAMIC_EXECUTION]: The skill guides the user through compiling DPA-side source code into executable binary images using the dpacc tool during the development process (TASKS.md).
  • [INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection attack surface by processing user-supplied source code. 1. Ingestion points: User-provided DPA kernel and host-side application code. 2. Boundary markers: No explicit instructions for delimiting untrusted code. 3. Capability inventory: Execution of dpacc, pkg-config, and the resulting binary. 4. Sanitization: No specific sanitization methods for the user code are mentioned.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Oct 15, 2026, 02:56 AM