verl-rl-training

Pass

Audited by Gen Agent Trust Hub on Jun 29, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill is a comprehensive technical guide for reinforcement learning training using the verl (Volcano Engine RL) library, an open-source framework.
  • [EXTERNAL_DOWNLOADS]: The skill provides instructions to download the verl library from its official GitHub repository (github.com/volcengine/verl) and use official Docker images. These are legitimate resources associated with the project's developers.
  • [COMMAND_EXECUTION]: The provided shell commands are standard for machine learning workflows, including environment setup (pip install), distributed computing management (ray start), and training execution (python -m verl.trainer.main_ppo).
  • [INDIRECT_PROMPT_INJECTION]: The skill defines patterns for processing training datasets and implementing reward functions. While this represents a data ingestion surface, it is consistent with the core functionality of a model training framework and contains no malicious directives.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 29, 2026, 12:55 AM
Security Audit — agent-trust-hub — verl-rl-training