huggingface-import
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external HuggingFace repositories, which are untrusted sources. If a dataset record contains malicious instructions, they could influence the agent during the import process or downstream evaluation.
- Ingestion points: Data is fetched from user-specified HuggingFace repositories, spaces, or datasets (SKILL.md, Step 1).
- Boundary markers: The instructions do not define clear delimiters or warnings to treat the imported data content as non-executable text.
- Capability inventory: The skill involves writing structured CSV files to the local file system (examples/huggingface-import.py) and making network requests to the Coval API at api.coval.dev (SKILL.md).
- Sanitization: The workflow focuses on structural formatting (CSV escaping and JSON metadata) but lacks mechanisms to sanitize or validate the actual content of the input fields against prompt injection patterns.
- [EXTERNAL_DOWNLOADS]: The skill fetches configuration and API specifications from api.coval.dev and accesses content from huggingface.co. These are expected behaviors for the skill's purpose of importing datasets to the vendor's platform.
- [COMMAND_EXECUTION]: The skill describes using shell commands (GET) to interact with the Coval API for fetching OpenAPI specifications.
Audit Metadata