The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides installation commands that fetch and execute shell scripts directly from official Hugging Face domains and GitHub repositories.
Evidence: curl -LsSf https://hf.co/cli/install.sh | bash -s and curl -fsSL https://raw.githubusercontent.com/huggingface/hf-mount/main/install.sh | sh.
[COMMAND_EXECUTION]: The skill facilitates the execution of shell commands and external extensions.
Evidence: Commands such as hf extensions install, hf jobs run, and hf sandbox exec allow the agent to run code or manage remote execution environments.
[DATA_EXFILTRATION]: The skill provides capabilities to manage and transmit sensitive data and credentials.
Evidence: It manages Hugging Face authentication tokens (HF_TOKEN) through commands like hf auth login and hf auth token. It also provides utilities for uploading local files and environment variables to remote Hugging Face buckets and repositories (hf upload, hf buckets cp).
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to its processing of untrusted external data.
Ingestion points: The skill reads model/dataset cards, discussion comments, and academic papers from the Hugging Face Hub (SKILL.md).
Boundary markers: None identified; external content is processed without explicit delimiters or instructions to ignore embedded commands.
Capability inventory: The skill possesses capabilities for file writing (hf download), command execution (hf sandbox exec), and network operations (hf upload).
Sanitization: There is no mention of sanitization or filtering of the content retrieved from the Hub before it is processed by the agent.

hf-cli