The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/generate_pdf.py uses the subprocess module to execute system commands for pandoc and xelatex. This is used to convert markdown documents into PDF files. While legitimate for document generation, it constitutes execution of external binary tools.
[EXTERNAL_DOWNLOADS]: The skill instructs users to install external system dependencies such as pandoc and MacTeX/TeX Live via package managers like brew and apt-get. Additionally, scripts/verify_citations.py makes network requests to well-known research services including doi.org and api.crossref.org to retrieve paper metadata.
[PROMPT_INJECTION]: The skill is susceptible to Indirect Prompt Injection (Category 8) due to the following evidence chain:
Ingestion points: Untrusted data from scientific databases (titles, abstracts, and metadata) is ingested into the agent's context through search results processed by scripts/search_databases.py.
Boundary markers: The skill does not implement explicit boundary markers or instructions for the agent to ignore potentially malicious content embedded within research paper abstracts or titles.
Capability inventory: The skill has capabilities including network access via requests in scripts/verify_citations.py and local command execution via subprocess in scripts/generate_pdf.py.
Sanitization: There is no evidence of sanitization or filtering of the external research content before it is synthesized into the final literature review document.

literature-review