The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references an installation command that downloads a script from Hugging Face's official domain (hf.co) and executes it via the shell. As this targets a well-known technology service, the reference is considered informative rather than malicious.
[COMMAND_EXECUTION]: The skill facilitates extensive interaction with the local file system and remote repositories through the hf CLI, including uploading, downloading, and synchronizing repository content.
[DYNAMIC_EXECUTION]: The documentation describes features for executing raw SQL queries against datasets and running Python scripts with inline dependencies using the uv tool, both of which involve runtime code evaluation.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface for indirect prompt injection by processing external data from the Hugging Face Hub.
Ingestion points: Untrusted data enters the context through repository files, dataset parquet metadata, hub discussions, and daily paper lists referenced in SKILL.md.
Boundary markers: The instructions lack explicit delimiters or warnings to ignore instructions embedded within the downloaded Hub content.
Capability inventory: The tool provides capabilities to execute SQL and Python jobs based on the ingested data.
Sanitization: No sanitization or validation logic is defined within the skill to mitigate the risk of instructions embedded in external assets.

huggingface-hub