The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted metadata (titles, authors) from the Semantic Scholar API. This creates a surface where a malicious paper title could influence the agent's behavior. Ingestion points: scripts/bibtex_fetch.py via search_s2 function. Boundary markers: Absent; metadata is interpolated directly into the BibTeX output. Capability inventory: The skill performs network operations to academic APIs. Sanitization: Basic whitespace cleaning is performed for formatting, but no filtering for prompt injection patterns is present.
[EXTERNAL_DOWNLOADS]: Fetches metadata from official arXiv and Semantic Scholar APIs.

bibtex-fetch