The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/generate_pdf.py utilizes subprocess.run to invoke the pandoc and xelatex binaries for document conversion. The command is constructed using a list of arguments rather than a raw shell string, which follows security best practices for preventing command injection.
[EXTERNAL_DOWNLOADS]: The scripts/verify_citations.py script performs network requests to academic infrastructure such as doi.org and the CrossRef API (api.crossref.org) to fetch paper metadata. These are official and well-known services for the scientific community.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and summarize data from external academic sources such as PubMed, arXiv, and bioRxiv. This creates a surface for indirect prompt injection if malicious content is present in the source metadata, although this risk is inherent to literature synthesis tools.
Ingestion points: Search result JSON files and API responses from academic databases are processed by scripts/search_databases.py and scripts/verify_citations.py.
Boundary markers: The project uses a structured markdown template (assets/review_template.md) to define where different types of content are placed.
Capability inventory: The skill is capable of file system writes (reports and PDFs) and executing specific document processing binaries via the shell.
Sanitization: No specific filtering or escaping logic for LLM-based instructions within fetched paper metadata was observed.

alterlab-literature-review