long-context

Pass

Audited by Gen Agent Trust Hub on Jun 29, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill provides legitimate machine learning implementation details and documentation for context extension techniques (RoPE, YaRN, ALiBi). No malicious patterns, obfuscation, or data exfiltration vectors were identified.
  • [EXTERNAL_DOWNLOADS]: The skill instructs the user to install standard Python packages from official registries (transformers, torch, einops, rotary-embedding-torch, flash-attn) and references a public research repository for YaRN. These downloads are appropriate and necessary for the stated technical purpose of the skill.
  • [COMMAND_EXECUTION]: Includes standard shell commands for installing dependencies and running training scripts, which is expected behavior for a technical machine learning development skill.
  • [SAFE]: Mentions of Hugging Face API usage for model management follow standard practices and do not involve hardcoded credentials or unauthorized access.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 29, 2026, 12:55 AM
Security Audit — agent-trust-hub — long-context