The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions to download installation scripts for the hf CLI and hf-mount tool from https://hf.co/cli/install.sh and https://raw.githubusercontent.com/huggingface/hf-mount/main/install.sh. These resources are hosted on Hugging Face's official infrastructure.
[REMOTE_CODE_EXECUTION]: The skill facilitates the execution of downloaded installation scripts and allows for running remote scripts on Hugging Face infrastructure via the hf jobs uv run and hf jobs scheduled uv run commands.
[COMMAND_EXECUTION]: The skill exposes a wide range of CLI commands for interacting with the Hugging Face Hub, including file transfers (hf cp, hf upload), repository management (hf repos create, hf repos delete), and environment configuration.
[DYNAMIC_EXECUTION]: Through the hf extensions and hf jobs subcommands, the skill can install extensions from GitHub repositories and execute arbitrary code or scripts within remote environments.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection by retrieving and processing third-party content from the Hugging Face Hub.
Ingestion points: Data enters the agent context through commands that read repository metadata, dataset cards, model cards, and research papers (e.g., hf datasets card, hf papers read).
Boundary markers: The skill instructions do not explicitly mandate the use of delimiters or specific instructions to the agent to ignore potential commands embedded in retrieved Hub content.
Capability inventory: The skill has significant capabilities, including persistent storage management, network exfiltration potential via uploads, and infrastructure deployment (e.g., hf endpoints deploy).
Sanitization: There are no documented procedures for sanitizing or filtering the content of dataset cards or papers before they are processed by the agent.

hugging-face-cli