The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the subprocess module to execute system commands such as pdflatex, bibtex, and chktex. These calls are localized to the scripts/assembly/ and scripts/formatting/ directories and are necessary for the core functionality of compiling and checking LaTeX documents.\n- [EXTERNAL_DOWNLOADS]: Several scripts (search_openalex.py, download_arxiv_source.py, etc.) perform network requests to well-known academic repositories and APIs. These are used to retrieve research papers, metadata, and LaTeX source code for literature reviews and citation harvesting.\n- [DYNAMIC_EXECUTION]: The stat_summary.py and figure_template.py scripts generate Python code based on user data and internal templates. This generated code is designed to be executed by the researcher for data analysis and visualization.\n- [INDIRECT_PROMPT_INJECTION]: The skill has a vulnerability surface for indirect prompt injection as it ingests untrusted data from external sources (e.g., arXiv source files and paper abstracts). \n
Ingestion points: External content enters the system via literature search and download scripts (e.g., scripts/literature/download_arxiv_source.py).\n
Boundary markers: The instructions do not define strict boundary markers or include warnings to ignore instructions embedded in the downloaded research content.\n
Capability inventory: The agent can perform file writes and execute shell commands through the assembly pipeline.\n
Sanitization: While clean_latex.py sanitizes LaTeX-specific special characters, there is no evidence of filtering for natural language instructions intended to manipulate the agent's behavior.

paper-writing