hugging-face-jobs
Pass
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONCREDENTIALS_UNSAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill uses the
hf_jobs()tool to execute arbitrary Python code and shell commands on Hugging Face's remote infrastructure. - [EXTERNAL_DOWNLOADS]: The skill fetches and executes Python scripts from Hugging Face's official GitHub repositories and the Hugging Face Hub.
- [DYNAMIC_EXECUTION]: The agent is instructed to generate Python scripts dynamically and pass them as strings to the
hf_jobstool for execution. - [INDIRECT_PROMPT_INJECTION]: The skill processes external data by ingesting datasets from the Hugging Face Hub for tasks like batch inference and synthetic data generation.
- Ingestion points: Loading datasets from the Hugging Face Hub (e.g.,
username/input-dataset). - Boundary markers: None mentioned in the script templates.
- Capability inventory: The
hf_jobstool provides capabilities for code execution, network operations, and writing results back to the Hub. - Sanitization: No explicit sanitization of dataset content is described before processing.
- [CREDENTIALS_UNSAFE]: The documentation includes examples of potentially unsafe credential management, such as hardcoding tokens in secrets or using environment variables, though it explicitly warns against these methods and recommends a secure automatic token replacement system.
Audit Metadata