huggingface-import

Pass

Audited by Gen Agent Trust Hub on Apr 9, 2026

Risk Level: SAFE
Full Analysis
  • [DATA_EXFILTRATION]: The skill documents interactions with https://api.coval.dev/v1, which is the official API for the Coval platform (the skill author). These operations are intended for fetching configuration and uploading test sets.
  • [COMMAND_EXECUTION]: The example script examples/huggingface-import.py uses the pathlib and csv modules to create directories and write files to the local filesystem. This is required to generate the output CSV files described in the workflow.
  • [EXTERNAL_DOWNLOADS]: The skill references several well-known datasets hosted on HuggingFace, such as cais/mmlu and openai/gsm8k. These are industry-standard benchmarks used for AI evaluation.
  • [PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection (Category 8) because it processes untrusted data from external sources.
  • Ingestion points: HuggingFace repositories, spaces, or datasets specified by the user or provided in $ARGUMENTS.
  • Boundary markers: None identified in the workflow or the generated CSV structure.
  • Capability inventory: The skill can write files to the local disk and send data to the Coval API.
  • Sanitization: The skill does not currently implement or instruct the user on sanitization of the imported dataset content before conversion.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 9, 2026, 11:43 AM