The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary workflow involves ingesting user-supplied research content (titles, text, and figures) and interpolating it into LaTeX templates. This process creates a surface for indirect prompt injection if the processed content contains malicious instructions designed to influence the agent during the writing or compilation phase.
Ingestion points: Research data and content outlined in the Stage 3 workflow of the SKILL.md file.
Boundary markers: The instructions do not define specific delimiters or guardrail instructions (e.g., 'ignore embedded commands') when handling user-provided strings for template population.
Capability inventory: The agent utilizes Bash for running compilation tools like pdflatex and the review_poster.sh script, along with Write capabilities to create the source files, which could be exploited if malicious LaTeX code is successfully injected.
Sanitization: No explicit sanitization or validation of the research content is performed before it is written to the LaTeX source files.

latex-posters