nemo-curator

Pass

Audited by Gen Agent Trust Hub on Oct 23, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [SAFE]: The skill provides instructions and code snippets for using the NVIDIA NeMo Curator library, which is an industry-standard toolkit for preparing training data.
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process large-scale external datasets (e.g., Common Crawl) as evidenced by the DocumentDataset.read_parquet calls in SKILL.md. This risk is mitigated by the skill's core functionality, which focuses on identifying and removing toxic content, PII, and low-quality data via modules like PIIRedactor and NSFWClassifier.
  • [EXTERNAL_DOWNLOADS]: The skill utilizes standard installation procedures for well-known Python packages (nemo-curator, cudf, dask) and references pre-trained models from established organizations such as NVIDIA and OpenAI, all of which are recognized as safe, well-known services.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 23, 2026, 07:23 PM