The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and summarize external content from Hugging Face paper pages and arXiv. This introduces an attack surface where malicious actors could embed instructions within research papers or paper metadata to influence agent behavior.
Ingestion points: Fetches content from https://huggingface.co/papers/{PAPER_ID}.md, https://huggingface.co/api/papers/{PAPER_ID}, and arXiv fallbacks.
Boundary markers: The skill instructions do not specify the use of delimiters or 'ignore' instructions when processing fetched markdown or JSON data.
Capability inventory: The agent can execute shell commands via curl, write/update paper links, and claim authorship using provided credentials.
Sanitization: No explicit sanitization or validation of the fetched paper content is mentioned before it is processed by the agent.
[COMMAND_EXECUTION]: The skill relies on shell curl commands to interact with external APIs. While these target well-known and trusted services, they incorporate variables like {PAPER_ID} and {ARXIV_ID} which originate from user input.
The instructions provide a regex pattern (^\d{4}\.\d{4,5}$) for the arxivId in the indexing endpoint, which serves as a security control against command injection.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill uses the $HF_TOKEN environment variable for authenticated API requests. This is a standard practice for managing credentials in agent environments and does not constitute a leak itself, provided the environment is secure.

huggingface-papers