The Agent Skills Directory

[SAFE]: No malicious patterns or security risks were identified during the analysis of the skill instructions and reference materials. The content is technically accurate and focused on reinforcement learning workflows.
[EXTERNAL_DOWNLOADS]: The skill references downloads from official and well-known infrastructure providers, such as the VolcEngine GitHub organization for source code and Docker Hub for container images. These are considered standard and legitimate sources for machine learning training tasks.
[COMMAND_EXECUTION]: The provided commands for environment initialization (pip, docker) and launching training jobs (Ray, Python) are routine for machine learning engineering and present no evidence of privilege escalation, persistence, or malicious intent.
[INDIRECT_PROMPT_INJECTION]: While the skill involves processing external datasets and custom reward functions, these ingestion surfaces are inherent to reinforcement learning training. The instructions provide structured ways to handle this data, aligning with best practices for development environments.

verl-rl-training