The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's instructions explicitly recommend using the -shell-escape flag with latexmk for specific document features (e.g., minted for code highlighting or TikZ externalization). This flag allows the LaTeX engine to execute arbitrary shell commands via the \write18 primitive.
[REMOTE_CODE_EXECUTION]: Enabling the shell-escape capability creates a vulnerability surface where a maliciously crafted LaTeX file could execute code on the host system during the compilation process. This risk is inherent to the primary function of advanced LaTeX typesetting but requires caution when processing external or unverified content.
[EXTERNAL_DOWNLOADS]: The skill facilitates the installation of dependencies from various standard sources. It uses npm to install the docx library, pip for markitdown (sourced from Microsoft's GitHub), and tlmgr for LaTeX packages from TeX Live repositories.
[PROMPT_INJECTION]: The skill processes user-supplied data (LaTeX source, bibliographies, and document content) through a multi-step compilation and conversion pipeline. It lacks explicit boundary markers or sanitization to prevent indirect prompt injection, where malicious instructions hidden in the document content could attempt to influence the agent's behavior during subsequent processing steps.
Ingestion points: Document content from m07/m08, reference data from m10, and figure metadata from m11 are interpolated into LaTeX and Word templates.
Boundary markers: None identified in the prompt interpolation logic.
Capability inventory: Subprocess execution of latexmk, tectonic, node, and python (via precheck_log.py).
Sanitization: No content escaping or validation logic is present before document generation.

light-typesetting