The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local system commands to process documents. The script scripts/check_float_pages.py uses subprocess.run to invoke the pdflatex command. Additionally, the instructions guide the agent to perform shell-based tasks using utilities like gs (Ghostscript) for PDF processing and sed for inline text editing.
[DYNAMIC_EXECUTION]: The scripts/check_float_pages.py script dynamically generates a modified LaTeX file (check.tex) in a temporary directory. It injects specific logging macros into the manuscript's source content before compiling it to extract page placement data for figures and tables.
[EXTERNAL_DOWNLOADS]: Reference materials in references/journal-switch.md recommend downloading the revtex package or other templates from the Comprehensive TeX Archive Network (CTAN), an established and trusted repository for LaTeX-related software.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted LaTeX source files which could potentially contain malicious macros. This ingestion of external data constitutes a surface for indirect prompt injection.
Ingestion points: scripts/check_float_pages.py and scripts/resample_figures.py read content directly from user-provided .tex files.
Boundary markers: The skill does not use specific delimiters or warnings to prevent the agent from following instructions that might be embedded in the LaTeX source.
Capability inventory: The skill has access to subprocess execution (via pdflatex), file system write permissions, and the ability to execute shell commands through the agent's environment.
Sanitization: No sanitization or validation of the LaTeX source code is performed prior to processing.

p-article-latex