hugging-face-jobs
Pass
Audited by Gen Agent Trust Hub on Aug 17, 2026
Risk Level: SAFEREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONCREDENTIALS_UNSAFEPROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill facilitates the generation and execution of arbitrary Python code and Docker containers on Hugging Face cloud infrastructure via the
hf_jobstool. This is the primary intended function of the skill. - [COMMAND_EXECUTION]: Uses MCP tools and CLI commands (e.g.,
hf jobs uv run,hf jobs logs) to interact with remote compute resources. - [EXTERNAL_DOWNLOADS]: Fetches scripts and assets from well-known services, specifically official Hugging Face GitHub repositories and the Hugging Face Hub (e.g.,
raw.githubusercontent.com/huggingface/*andhuggingface.co/datasets/uv-scripts/*). - [CREDENTIALS_UNSAFE]: Provides explicit guidance on secure authentication using the platform's native secret replacement mechanism (
$HF_TOKEN) and includes warnings against hardcoding sensitive tokens in source code. - [DATA_EXFILTRATION]: Documents methods for persisting results to the Hugging Face Hub or external storage providers (AWS S3, GCS) as a required part of the ephemeral job lifecycle.
- [PROMPT_INJECTION]: The skill documents workflows for processing untrusted datasets from the Hugging Face Hub (e.g.,
username/input-datasetin fileSKILL.md). - Ingestion points: Reads dataset content from the Hugging Face Hub.
- Boundary markers: None explicitly defined in the provided instruction templates.
- Capability inventory: Remote script execution, network access, and repository write access via the
hf_jobstool. - Sanitization: No specific sanitization or filtering logic is described for the processed data.
Audit Metadata