The Agent Skills Directory

[SAFE]: The skill provides documentation and code snippets for using the NeMo Curator library for data processing. All dependencies listed (nemo-curator, cudf, dask, rapids) are standard libraries for GPU data science and LLM training data preparation.
[EXTERNAL_DOWNLOADS]: The skill mentions installation commands using uv pip install for nemo-curator. These target official package registries and are standard for the described use case. It also references official models on Hugging Face (e.g., nvidia/quality-classifier-deberta, sentence-transformers/all-MiniLM-L6-v2) which are trusted sources for machine learning artifacts.
[COMMAND_EXECUTION]: Shell commands are restricted to package installation (uv pip install). Python code examples demonstrate data manipulation and processing using the library's official API. No suspicious subprocess calls or arbitrary command executions were found.
[DATA_EXFILTRATION]: No evidence of sensitive data access or unauthorized network exfiltration. The network activity is limited to downloading official models and reading datasets from user-defined paths (e.g., S3 or local Parquet files), which is consistent with the skill's purpose.

nemo-curator