dataset-curation

Pass

Audited by Gen Agent Trust Hub on Jul 14, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The instructions focus on best practices for preparing machine learning datasets, including format selection, template application, and loss masking to ensure training stability and quality.\n- [SAFE]: Python code examples are provided for data inspection and format conversion; these snippets perform standard operations using well-known libraries such as transformers and trl.\n- [SAFE]: The skill includes a robust synthetic data management framework with filtering funnels (deduplication, quality scoring) and minimum real-data thresholds to protect against model collapse.\n- [SAFE]: No obfuscation, data exfiltration patterns, or unauthorized remote code execution were detected in the instructions or scripts.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 14, 2026, 07:33 PM
Security Audit — agent-trust-hub — dataset-curation