The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill documentation in references/using-models.md describes loading scientific models from the Hugging Face Hub using the trust_remote_code=True flag. This is noted as a requirement for custom architectures like Evo-2 and single-cell foundation models. The instructions correctly require the agent to inform the user that this flag executes custom modeling code from the repository, which is standard procedure for these specialized models on the Hugging Face platform.
[EXTERNAL_DOWNLOADS]: The scripts/fetch_catalog.py script fetches markdown resource indices and topic files from the external domain huggingscience.co. This domain serves as the primary data source for the skill's curation functionality.
[COMMAND_EXECUTION]: The skill includes a bundled Python script, scripts/fetch_catalog.py, which the agent uses to perform structured queries against the catalog. The script uses standard libraries and performs simple network operations to retrieve catalog metadata.
[INDIRECT_PROMPT_INJECTION]: The skill ingest markdown content from an external discovery catalog. This creates a surface where external content could influence agent behavior. However, this is the core intended functionality of the skill, and the data is parsed into structured entries (type, tags, URL, description) rather than being directly executed as instructions.

hugging-science