The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the requests Python library and standard system utilities such as pandoc and xelatex for document processing and PDF generation. These are reputable tools commonly used in academic and technical environments.
[COMMAND_EXECUTION]: The script scripts/generate_pdf.py automates PDF creation by executing pandoc and xelatex through subprocess.run. The implementation is secure, as it passes arguments as a list and avoids the use of shell=True, effectively preventing command injection risks from user-supplied file names or paths.
[DATA_EXFILTRATION]: While the skill performs network requests via scripts/verify_citations.py, these operations are strictly confined to official citation metadata services (doi.org and crossref.org). This is a legitimate and necessary function for ensuring the accuracy of research citations.
[INDIRECT_PROMPT_INJECTION]: The skill features an indirect prompt injection surface as it processes untrusted data from academic databases (e.g., PubMed, arXiv).
Ingestion points: External citation data and abstracts are ingested via JSON files and processed by scripts/search_databases.py.
Boundary markers: The instructions do not define specific markers to isolate the untrusted search results from the agent's instructions.
Capability inventory: The skill has access to the Bash tool and executes shell commands via the PDF generation script.
Sanitization: There is no explicit sanitization of external metadata before it is interpolated into the final review documents. This is considered a standard operational risk for research-oriented skills.
[SAFE]: No evidence of prompt injection, hidden persistence mechanisms, or credential harvesting was found. The skill maintains a professional, academic focus and includes verification steps that promote data integrity.

literature-review