The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external academic sources, creating a potential surface for indirect prompt injection attacks.\n
Ingestion points: fetch_paper.py fetches full paper text from external academic URLs (e.g., arxiv.org, doi.org) and the Jina Reader service; multiple scripts fetch paper titles, abstracts, and TLDRs from Semantic Scholar and DeepXiv/arXiv APIs.\n
Boundary markers: The instructions do not define explicit delimiters or boundary markers to isolate external paper content from agent instructions.\n
Capability inventory: The skill is granted write_file to save papers/notes and execute to run the provided Python scripts.\n
Sanitization: fetch_paper.py implements _strip_boilerplate to remove website navigation and cookie consent text from fetched content before it is processed by the agent.\n- [EXTERNAL_DOWNLOADS]: The skill facilitates downloading content and metadata from external academic platforms and utility services.\n
download_paper.py downloads PDF files from arxiv.org and Semantic Scholar's open-access URLs.\n
fetch_paper.py retrieves paper content using the r.jina.ai service.\n
Metadata and search results are retrieved from well-known academic APIs including Semantic Scholar, arXiv (via DeepXiv), HuggingFace, and GitHub.

paper-navigator