verl-rl-training
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill serves as a technical manual and configuration guide for reinforcement learning training workflows. All instructions and code snippets are standard for the machine learning domain.
- [EXTERNAL_DOWNLOADS]: The skill references installation of the 'verl' library via pip, Docker, and GitHub. These resources originate from the official repositories of Volcengine (ByteDance) and the verl-project, which are recognized sources for this technology. The dependencies listed (torch, ray, vllm, transformers) are standard industry-standard libraries.
- [COMMAND_EXECUTION]: The shell commands provided are for launching distributed training jobs (using Ray) and installing necessary packages. These commands are transparent, do not contain obfuscation, and are intended for execution by a user in a controlled environment.
- [DATA_EXFILTRATION]: No patterns of data exfiltration or unauthorized file access were found. The skill references local datasets (e.g., parquet files) as part of the training data ingestion process, which is the expected behavior for a training utility.
Audit Metadata