cuda-skill
Pass
Audited by Gen Agent Trust Hub on Mar 4, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: No attempts to override agent behavior or bypass safety guidelines were found in the skill instructions or the extensive documentation files.
- [DATA_EXFILTRATION]: Access is restricted to the skill's own documentation directory. No access to sensitive system paths or unauthorized network operations was detected.
- [COMMAND_EXECUTION]: The skill utilizes standard search and development tools such as ripgrep (rg), cat, and the CUDA toolkit (nvcc, nsys, ncu) as intended for its purpose as a development assistant. These commands are provided as templates for the user to execute on their own source code.
- [REMOTE_CODE_EXECUTION]: No remote code execution patterns or untrusted external downloads were found. The documentation update scripts mentioned rely on well-known and trusted technology repositories.
- [INDIRECT_PROMPT_INJECTION]: While the skill processes a large volume of local documentation files, the risk is negligible as the data consists of standard technical specifications and the search strategy is designed to retrieve specific fragments rather than processing untrusted data at runtime.
Audit Metadata