The Agent Skills Directory

[DYNAMIC_EXECUTION]: The evaluation script evals/files/cudf-native-stream-handoff-boundary/code/run_smoke.sh uses the nvcc compiler to build the threaded_handoff.cu CUDA C++ file at runtime to verify GPU stream handoff logic.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it is designed to ingest and process external data files (CSV, Parquet, JSON) which could contain malicious instructions.
Ingestion points: read_csv, read_parquet, and read_json operations across multiple scripts including etl_pipeline.py, parquet_pipeline.py, and null_pipeline.py.
Boundary markers: None detected.
Capability inventory: File system writes (to_parquet, to_csv), directory deletion (shutil.rmtree in parquet_pipeline.py), and shell command execution (python -m cudf.pandas).
Sanitization: Standard data processing occurs without specific natural language instruction filtering.
[EXTERNAL_DOWNLOADS]: The skill contains references to external documentation and source code repositories on official RAPIDS and NVIDIA domains, such as docs.rapids.ai and github.com/rapidsai.
[COMMAND_EXECUTION]: The documentation provides instructions for executing Python scripts with the cudf.pandas accelerator module using python -m cudf.pandas.

accelerated-computing-cudf