skills/huggingface/skills/hf-cli/Gen Agent Trust Hub

hf-cli

Pass

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: SAFEREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • Remote Installation Pattern: The skill documents an installation method using curl piped to bash from the vendor's official domain (hf.co). This is a common pattern for distributing developer tools and is used here to set up the official CLI.
  • Remote Execution Capabilities: The command set includes functions like hf endpoints deploy, hf jobs run, and hf sandbox exec, which enable the management and execution of code on Hugging Face's remote cloud infrastructure. These features are designed for managing AI training and inference workflows.
  • Extension Management: The hf extensions install command allows the installation of additional tools from GitHub repositories. This provides extensibility for the CLI, though the security of specific extensions depends on the external repository source.
  • Authentication Handling: Commands under hf auth allow for managing Hugging Face access tokens. This is a core functionality required to authenticate with the Hub for accessing private repositories and managing account settings.
  • Indirect Prompt Injection Surface: The skill provides tools to read external content from the Hub, such as dataset cards, model cards, and research papers. This information is ingested into the agent's context to provide information to the user, a standard retrieval pattern that represents a common ingestion surface.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 15, 2026, 01:01 AM
Security Audit — agent-trust-hub — hf-cli