The Agent Skills Directory

[SAFE]: The skill serves as a comprehensive guide for Ray Train. All external references target official and well-known resources, such as the Ray project's GitHub, official documentation (docs.ray.io), and standard package registries (PyPI, Helm).
[EXTERNAL_DOWNLOADS]: The instructions include standard installation commands (pip install -U ray[train]) and configuration for well-known cloud providers (AWS, GCP, Azure) and Kubernetes. These are standard operations for setting up a distributed training environment.
[INDIRECT_PROMPT_INJECTION]: As an ML training orchestration skill, it naturally ingests configuration dictionaries and datasets. While this represents a potential surface for processing untrusted data, the skill follows standard library patterns for Ray and PyTorch, and the capability to execute code across workers is the primary, legitimate purpose of the tool.
[COMMAND_EXECUTION]: The skill documents the use of CLI tools like ray start, ray up, and ray job submit. These are intended for cluster management and are used in accordance with their official documentation.
[DATA_EXFILTRATION]: No patterns of unauthorized data access or sensitive file exfiltration were detected. The use of S3 URLs (s3://my-bucket/ray-results) is documented for legitimate training result storage.

ray-train