The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions include numerous curl command examples used to interact with the Hugging Face API and fetch content from arXiv.
[EXTERNAL_DOWNLOADS]: The skill fetches paper content and metadata from huggingface.co and arxiv.org. These are well-known services in the AI research community.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) because it ingests untrusted markdown content and metadata from research papers which could contain adversarial instructions.
Ingestion points: Paper content fetched from huggingface.co/papers/{PAPER_ID}.md and metadata from huggingface.co/api/papers/{PAPER_ID}.
Boundary markers: None identified in the skill instructions to distinguish between paper content and agent instructions.
Capability inventory: The skill uses curl for network operations and processes text for summarization and analysis.
Sanitization: The skill does not specify any sanitization or filtering of the ingested paper content.

huggingface-papers