The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user research data and interpolates it into LaTeX templates. If the agent does not properly escape LaTeX special characters or if the data contains malicious LaTeX commands targeting shell escape functionalities (like \write18), this could lead to unintended command execution during PDF compilation.\n- Ingestion points: User research content is processed and integrated into templates during the Stage 3 workflow defined in SKILL.md.\n- Boundary markers: Explicit boundary markers or instructions to escape user-supplied data are absent from the interpolation process.\n- Capability inventory: The skill possesses Bash access, Write permissions, and the capability to execute pdflatex on generated files.\n- Sanitization: No explicit sanitization or escaping of LaTeX special characters is performed on user input.\n- [EXTERNAL_DOWNLOADS]: The skill documentation in SKILL.md suggests using the tlmgr package manager to install necessary LaTeX packages from the official TeX Live repository.\n- [COMMAND_EXECUTION]: The skill utilizes Bash permissions to execute PDF compilation commands such as pdflatex and runs an automated review script review_poster.sh. The SKILL.md file also references a missing local script scripts/generate_schematic.py for mandatory figure generation.\n- [DYNAMIC_EXECUTION]: The skill generates LaTeX source files from templates and executes compilation engines (pdflatex, lualatex, xelatex) at runtime to produce PDF outputs.

latex-posters