The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill instructions direct the agent to use sudo for administrative tasks, specifically installing system packages via apt on Linux and tlmgr on macOS.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface for processing untrusted data.
Ingestion points: Content is read from user-provided .tex files in Step 2 to extract document classes and required packages.
Boundary markers: Absent. The skill does not provide instructions to delimit or ignore potentially malicious content within the LaTeX files.
Capability inventory: The skill has the capability to execute shell commands, install system-wide packages via sudo, and run local compilers.
Sanitization: Absent. There are no validation steps to ensure that extracted package names or configuration settings are safe before being used in shell commands.
[COMMAND_EXECUTION]: The skill uses various shell commands for environment detection (which, pdflatex --version), package management (tlmgr install), and multi-stage document compilation pipelines.
[DYNAMIC_EXECUTION]: The skill references the write18 feature in Step 5. If shell-escape (write18) is enabled during compilation, the LaTeX compiler can execute arbitrary shell commands defined within the document. Processing a malicious or untrusted .tex file with this feature enabled could result in unauthorized command execution on the host system.
[EXTERNAL_DOWNLOADS]: The skill encourages the agent to download venue-specific templates from external web sources. While the instructions specify official sources, this pattern involves fetching and processing remote content that could be spoofed or compromised.

latex-setup