The Agent Skills Directory

[PROMPT_INJECTION]: Indirect Prompt Injection Surface.
Ingestion points: The scripts scripts/precheck_log.py and scripts/submission_check.py ingest untrusted data from LaTeX log files, .tex source files, and .pdf files.
Boundary markers: There are no explicit boundary markers or instructions to ignore embedded commands when the agent processes the output of these validation scripts.
Capability inventory: The skill environment allows execution of shell commands via latexmk, python, and node.
Sanitization: The scripts use regular expressions to extract specific patterns but do not perform comprehensive sanitization to prevent the extracted text from containing instructions that might bypass agent safeguards.
[COMMAND_EXECUTION]: The skill facilitates the execution of local scripts and typesetting engines.
The Python scripts scripts/precheck_log.py and scripts/submission_check.py are used to validate document compliance and compilation success.
The Node.js script assets/docx_template.js is executed to generate Word documents using the docx library.
latexmk is used to automate the LaTeX compilation process, including the use of -shell-escape for specific packages like minted.
[EXTERNAL_DOWNLOADS]: The skill references several well-known and trusted external sources for LaTeX templates.
Instructions guide the user to download official templates from publisher websites (e.g., IEEE, ACM), Overleaf's gallery, and CTAN.
TeX package management is performed via tlmgr and tinytex, which fetch packages from official TeX repositories.

light-typesetting