accelerated-computing-cudf
Pass
Audited by Gen Agent Trust Hub on Jun 18, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill instructions and associated reference files focus exclusively on technical documentation and best practices for the RAPIDS ecosystem. No malicious behaviors were identified.
- [PROMPT_INJECTION]: The skill uses structured 'Critical Rules' to guide the agent toward efficient GPU usage (e.g., row count thresholds), which are legitimate operational constraints and do not attempt to bypass safety filters.
- [DATA_EXPOSURE]: Data handling is restricted to common ETL file formats and generated synthetic datasets. There is no evidence of hardcoded credentials or access to sensitive system configuration files.
- [COMMAND_EXECUTION]: The skill documents standard development commands such as profiling with the 'cudf.pandas' module and compiling CUDA code with 'nvcc' in the evaluation examples, all of which are appropriate for its intended use case.
- [REMOTE_CODE_EXECUTION]: No patterns of downloading and piping remote scripts to shells (e.g., curl|bash) were found. References to remote S3 paths are limited to data ingestion patterns.
Audit Metadata