skills/a5c-ai/babysitter/cuda-graphs/Gen Agent Trust Hub

cuda-graphs

Pass

Audited by Gen Agent Trust Hub on Apr 1, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: No malicious patterns or security risks were identified. The skill's architecture is focused on performance optimization for GPU workloads using standard CUDA runtime calls and profiling tools (Nsight Systems).\n- [SAFE]: The skill requests Bash and file system permissions which are used legitimately to execute toolchain commands (e.g., DOT for visualization, benchmarking scripts) and to manage CUDA source files and binary artifacts.\n- [SAFE]: Analysis of the input processing surface (Category 8) shows that the skill reads CUDA source files and graph binaries for optimization. This is the primary intended purpose of the skill and no suspicious prompt injection or data exfiltration patterns were found within the instructions or provided templates. (Ingestion points: source code and binary graph files; Boundary markers: Absent; Capability inventory: Bash, Read, Write, Edit; Sanitization: Standard file processing).
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 1, 2026, 07:31 AM