The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The SKILL.md file contains multiple instructions to use the 'sudo' command for system-level operations, such as updating TeX Live ('sudo tlmgr update --self'), installing packages ('sudo tlmgr install tcolorbox tikz pgf'), and copying files to system directories ('sudo cp assets/medical_treatment_plan.sty /usr/local/texlive/...'). This explicitly encourages the agent to acquire administrative privileges to modify the host environment.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted patient data and interpolates it into LaTeX templates (e.g., assets/general_medical_treatment_plan.tex). Because the skill uses the 'Bash' tool to compile these documents (via pdflatex or xelatex), it creates a vulnerability surface where malicious user input could potentially exploit LaTeX macro execution (like '\write18') to run arbitrary commands.
[COMMAND_EXECUTION]: The skill documentation and SKILL.md describe extensive use of the 'Bash' tool to execute Python scripts for validation ('python validate_treatment_plan.py'), template generation, and visual timeline creation, as well as invoking LaTeX compilers directly. The reliance on shell execution for processing user-generated files increases the risk profile of the skill.

treatment-plans