skills/peachest/skills/hf-download/Gen Agent Trust Hub

hf-download

Pass

Audited by Gen Agent Trust Hub on Oct 16, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill includes shell scripts (hfd and check-env.sh) that orchestrate downloads using curl, aria2c, and wget. These scripts handle network operations and local file management to facilitate the download process. The script implementation follows standard shell scripting practices for CLI tools, including handling environment variables for configuration.
  • [INDIRECT_PROMPT_INJECTION]: The hfd script processes metadata retrieved from the Hugging Face API to identify and download specific files. While processing data from an external source (Hugging Face) is an attack surface, the script uses robust parsing methods like jq to handle the JSON responses.
  • Ingestion points: Metadata is fetched from huggingface.co/api/ and stored in a temporary JSON file (.hfd/repo_metadata.json).
  • Boundary markers: None. The script relies on JSON structure and defined API endpoints.
  • Capability inventory: The script performs file system writes and initiates network downloads via subprocess calls to curl, aria2c, or wget.
  • Sanitization: File lists are generated using jq filters or regex-based extraction with grep and awk to ensure only specific fields from the metadata are used.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 16, 2026, 10:43 AM