The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of external data from web pages and PDF documents (Step 5: Deep dive into key papers). This creates a potential surface where the agent might process hidden instructions embedded in scientific content from the web.
Ingestion points: SKILL.md describes methods for extracting text from URLs using trafilatura and PDFs using pypdf.
Boundary markers: The provided code snippets do not include explicit delimiter-based isolation for the extracted content.
Capability inventory: The skill leverages network capabilities via requests and file system access to write supplementary materials.
Sanitization: No explicit sanitization or instruction-filtering logic is present for the retrieved text.
[EXTERNAL_DOWNLOADS]: The skill instructions and code examples reference fetching data and metadata from well-known and reputable scientific sources.
Evidence: Mentions and provides code for interacting with api.crossref.org, arxiv.org, nature.com, and pubmed.ncbi.nlm.nih.gov using official APIs or standard libraries.

scientific-literature-search