The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/generate_pdf.py utilizes subprocess.run() to invoke the pandoc system utility for converting markdown to PDF. While this is a core functionality of the skill, it involves executing system binaries that can be susceptible to command injection or unauthorized file access if the input markdown content contains malicious LaTeX directives (e.g., using \write18 or \input) that bypass standard sanitization.
[PROMPT_INJECTION]: The skill workflow involves ingesting and synthesizing data from external, untrusted scientific literature databases such as PubMed, bioRxiv, and arXiv. This creates a surface for Indirect Prompt Injection, where an attacker could publish research metadata (titles, abstracts) containing instructions intended to manipulate the agent's behavior during the synthesis phase. The analysis identified ingestion points in scripts/search_databases.py that process this data without explicit semantic boundary markers or robust sanitization.
[EXTERNAL_DOWNLOADS]: The scripts/verify_citations.py script performs network requests to well-known academic services doi.org and api.crossref.org to retrieve publication metadata. These operations are consistent with the skill's stated purpose of citation verification and use official API endpoints for scholarly communication.

literature-review