The Agent Skills Directory

[COMMAND_EXECUTION]: The script validate_presentation.py uses subprocess.run to execute the pdflatex command. This is used for its intended purpose of validating and compiling LaTeX Beamer templates provided in the skill's assets. The script also uses sys.executable in generate_slide_image.py to call internal generation modules.
[EXTERNAL_DOWNLOADS]: The generate_slide_image_ai.py script communicates with https://openrouter.ai/api/v1 to interface with AI models (Gemini) for slide generation and quality review. OpenRouter is a well-known service for LLM aggregation, and its use here is consistent with the skill's primary functionality.
[PROMPT_INJECTION]: The skill processes potentially untrusted data by ingesting user-provided images, charts, and figures (via the --attach flag in generate_slide_image.py) and incorporating them into prompts for Vision LLMs. This represents an indirect prompt injection surface.
Ingestion points: User-provided image attachments in generate_slide_image_ai.py; input PDF, PPTX, and TEX files in validate_presentation.py.
Boundary markers: None identified in the prompt construction logic in generate_slide_image_ai.py.
Capability inventory: Subprocess execution for template validation (pdflatex) and network access for AI requests (requests).
Sanitization: No explicit sanitization or filtering is performed on content extracted from ingested files.

scientific-slides