ray-data

Fail

Audited by Gen Agent Trust Hub on Feb 16, 2026

Risk Level: HIGHPROMPT_INJECTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATION
Full Analysis
  • [Indirect Prompt Injection] (HIGH): The skill possesses a significant attack surface for indirect prompt injection. Evidence:
  • Ingestion points: Data is loaded from untrusted external sources in SKILL.md (e.g., read_parquet, read_csv, read_json).
  • Boundary markers: No delimiters or protective instructions are provided to separate data from instructions.
  • Capability inventory: Powerful data transformation and output capabilities are available (e.g., map_batches, write_parquet).
  • Sanitization: No input validation or sanitization is implemented.
  • [External Downloads] (LOW): The skill references external Python packages (ray[data], pyarrow, pandas) for installation via pip. While these are reputable libraries, they represent external code dependencies.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 16, 2026, 05:19 AM