doca-gpunetio-ib-write-bw
Pass
Audited by Gen Agent Trust Hub on Jul 29, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: No malicious patterns or security vulnerabilities were detected in the skill instructions, scripts, or metadata. The skill focuses on providing technical guidance for the DOCA GPUNetIO benchmark.
- [EXTERNAL_DOWNLOADS]: The skill references the DOCA SDK and CUDA Toolkit as prerequisites. All provided URLs and repository references target official NVIDIA domains (nvidia.com) and GitHub organizations (NVIDIA-DOCA), which are trusted sources for this skill's context.
- [COMMAND_EXECUTION]: The skill documents standard build workflows using meson and meson compile to build benchmark binaries from local source code. These commands are necessary for the tool's operation and are documented alongside clear preconditions.
- [PROMPT_INJECTION]: The skill includes explicit instructions for the AI agent to avoid hallucinating flags or paths, referred to as a 'Do-not-invent guard', which enhances the reliability of the agent's output.
- [DATA_EXFILTRATION]: The skill explicitly advises redacting sensitive identifiers such as GPU-side handles, memory-mmap exports, and OOB connection descriptors from process output before they are persisted or shared, demonstrating a commitment to secure data handling practices.
Audit Metadata