The Agent Skills Directory

[DYNAMIC_EXECUTION]: The evaluation script run_smoke.sh invokes the nvcc compiler to build the threaded_handoff.cu source file at runtime. This process is used to verify the correctness of asynchronous GPU memory operations.\n- [INDIRECT_PROMPT_INJECTION]: Multiple scripts within the skill (e.g., etl_pipeline.py, null_pipeline.py) are designed to ingest data from CSV and Parquet files. This creates an attack surface where untrusted data could potentially influence the agent's context.\n
Ingestion points: Data ingestion occurs via pd.read_csv and pd.read_parquet in files like etl_pipeline.py and parquet_pipeline.py.\n
Boundary markers: There are no explicit boundary markers or instructions to ignore instructions embedded within the processed data files.\n
Capability inventory: The environment allows file system writes and command execution via shell scripts and compilers.\n
Sanitization: No input validation or sanitization is performed on the data fields during the ingestion process.\n- [COMMAND_EXECUTION]: The skill documentation and evaluation logic involve executing shell commands and Python CLI modules for performance profiling and code validation.

accelerated-computing-cudf