The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied Markdown files and converts them into LaTeX code. It follows a specific mapping for headings and formatting but allows math blocks ($...$ and $$...$$) to pass through unchanged. This creates a surface where malicious LaTeX commands could be embedded in the source draft and carried into the final compilation. However, the risk is inherent to the LaTeX toolchain and not an intentional vulnerability in the skill.
Ingestion points: Processes content from a Markdown file path provided as an argument.
Boundary markers: No specific boundary markers or 'ignore' instructions are used for the external content.
Capability inventory: Utilizes Write to create .tex files and Bash(latexmk*) to execute the compilation process.
Sanitization: The skill uses a deterministic conversion table which limits the translation of most Markdown syntax, though raw LaTeX within math blocks is preserved.
[COMMAND_EXECUTION]: The skill uses Bash to run latexmk, mkdir, and ls. These commands are scoped to building the document and managing the output directory (out/), which is standard behavior for LaTeX automation.
[DATA_EXPOSURE]: The skill includes instructions for bibliography management via Paperpile and references a local linting script scripts/bib/citation_lint.py. No credentials or sensitive environment data are accessed or transmitted externally.

latex-scaffold