The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions to download the verl library from its official GitHub repository and to pull official Docker images for production environments.
[COMMAND_EXECUTION]: Contains standard shell commands for installing Python dependencies, initializing distributed compute clusters via Ray, and launching training jobs using the framework's trainer modules.
[DYNAMIC_EXECUTION]: Documents the ability to load and execute custom reward functions from local Python files, which is a core and expected feature for reinforcement learning training pipelines.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external datasets for training purposes; while this involves ingesting untrusted data into the training pipeline, it is the primary intended function of the library and follows standard ML engineering practices.

verl-rl-training