skills/nvidia/omniperf/diagnose-perf/Gen Agent Trust Hub

diagnose-perf

Pass

Audited by Gen Agent Trust Hub on Jun 22, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses sudo cpupower to set the CPU frequency governor to 'performance'. This is a standard optimization for simulation tasks but requires administrative privileges to modify system hardware states.
  • [COMMAND_EXECUTION]: The skill executes several diagnostic and monitoring commands including nvidia-smi, top, free, lscpu, and pgrep to gather hardware and process metrics.
  • [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting and parsing performance benchmark JSON files.
  • Ingestion points: Benchmark JSON files located in /tmp/benchmark_output/*.json are read and processed (SKILL.md).
  • Boundary markers: No specific delimiters or instructions are used to encapsulate the external JSON content or warn the agent about embedded instructions.
  • Capability inventory: The skill context includes access to privileged system commands (sudo) and process monitoring tools.
  • Sanitization: The data is processed using Python's json.load() library, which is a safe parsing method but does not validate the schema or sanitize the measurement values before output.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 22, 2026, 06:55 AM
Security Audit — agent-trust-hub — diagnose-perf