The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill documentation and README files suggest the installation of standard Python libraries (bibtexparser, requests, semanticscholar, arxiv) from official package registries to enable the citation verification scripts.
[COMMAND_EXECUTION]: The provided Python scripts (api-clients.py, verify-citations.py) execute network operations using the requests library to fetch metadata from well-known scholarly APIs, including api.crossref.org, doi.org, api.semanticscholar.org, and export.arxiv.org. These operations are transparent, rate-limited, and limited to academic data retrieval.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it ingests and processes untrusted bibliographic data from .bib and .tex files.
Ingestion points: The scripts verify-citations.py and format-checker.py read user-provided BibTeX and LaTeX files.
Boundary markers: No explicit boundary markers or "ignore embedded instructions" warnings are applied to the data during the parsing and reporting phase.
Capability inventory: The skill scripts have the capability to perform network requests (requests) and read local files.
Sanitization: The skill uses bibtexparser for structured data extraction and regex-based normalization for matching, but does not provide specific protection against natural language instructions embedded within citation fields (e.g., in a paper's title or note field).

reference-audit-guide