cupynumeric-parallel-data-load
Pass
Audited by Gen Agent Trust Hub on Jun 18, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill implements legitimate high-performance computing patterns for distributed data loading using the NVIDIA Legate and cuPyNumeric frameworks.
- [DATA_EXPOSURE]: The implementation reads from local filesystem paths provided by the user. It uses memory-mapped access (
mmap_mode="r") to inspect file headers and determine array metadata, which is standard behavior for its stated purpose and does not involve external data transmission. - [COMMAND_EXECUTION]: The example script
assets/examples/parallel_npy_load.pyincludes a 'clean' command that utilizesshutil.rmtreeto remove the temporary data directory created during the demo. This is a transparent administrative function documented in the code. - [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied directory paths and file headers. It incorporates validation checks for naming conventions, data types, and array shapes to ensure data consistency before processing. Ingestion points: Directory paths and
.npyfile headers. Boundary markers: None (uses standard filesystem paths). Capability inventory: File read/write and directory management via Python and Legate APIs. Sanitization: Validation of file existence, naming sequences, and metadata (dtype/shape) matches.
Audit Metadata