nemo-curator
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to curate datasets from untrusted sources like web scrapes, which represents a potential surface for indirect prompt injection. This is mitigated by the skill's primary function of quality filtering and PII redaction.\n
- Ingestion points:
DocumentDataset.read_parquetin SKILL.md.\n - Boundary markers: None explicitly mentioned for untrusted content.\n
- Capability inventory: File writes via
to_parquetand S3 loading.\n - Sanitization: Includes specific modules for quality filtering, PII redaction, and NSFW detection.\n- [EXTERNAL_DOWNLOADS]: The skill fetches pre-trained models from well-known services including Hugging Face (e.g.,
sentence-transformers,openai,nvidia). These are standard resources for machine learning workflows.\n- [COMMAND_EXECUTION]: The skill provides standard installation commands for thenemo-curatorlibrary and its components.
Audit Metadata