The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines a surface area for processing untrusted data from external cloud storage, which is a known vector for indirect prompt injection.
Ingestion points: The skill utilizes ray.data.read_parquet, ray.data.read_csv, ray.data.read_json, and ray.data.read_images in SKILL.md to load datasets from external paths.
Boundary markers: Absent; there are no delimiters or specific instructions to ignore embedded commands within the processed data.
Capability inventory: The skill performs distributed transformations and writes to storage; it does not invoke shell commands, spawn subprocesses, or utilize dynamic execution tools like eval or exec on data content.
Sanitization: No data validation or escaping mechanisms are present in the provided transformation examples.
[EXTERNAL_DOWNLOADS]: The skill facilitates the installation of standard data science libraries and references established resources.
Installs ray[data], pyarrow, and pandas from official package registries.
References official documentation at docs.ray.io and the ray-project GitHub repository, which are well-known technology services.

ray-data