The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches citation metadata from established and trusted academic services, including CrossRef, NCBI/PubMed E-utilities, and the arXiv API. These operations are essential for the skill's primary purpose of academic reference management.
[COMMAND_EXECUTION]: Provides a suite of Python scripts (search_pubmed.py, extract_metadata.py, doi_to_bibtex.py, etc.) designed to automate research workflows. These scripts perform routine tasks such as network requests to academic APIs and writing formatted BibTeX data to the local file system.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data originating from external academic search results (e.g., paper titles and abstracts from Google Scholar or PubMed). This content is ingested and interpolated into the agent's context for synthesis and bibliography generation. The scripts provide structural parsing (XML/JSON) but do not include specific sanitization logic for potential instructions embedded within paper metadata.
[METADATA_POISONING]: The skill documentation includes specific instructions to adopt a certain persona or use external tools (e.g., "Nano Banana Pro" for schematic generation). While these instructions enforce specific workflow behaviors, they do not appear to bypass safety protocols or conceal malicious intent.

citation-management