The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes several scripts (scripts/generate_schematic.py, scripts/generate_slide_image.py, and scripts/validate_presentation.py) that utilize the subprocess.run function. These implementations are secure as they pass commands as argument lists rather than shell strings, avoiding shell injection risks. They are used appropriately to chain internal worker scripts or call standard local tools like pdflatex for document validation.
[INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection because it processes user-provided descriptions and attached images through an external LLM (Gemini via OpenRouter) to generate slide content.
Ingestion points: User-supplied prompts and image attachments provided to the generate_slide_image_ai.py and generate_schematic_ai.py scripts.
Boundary markers: The skill does not employ explicit delimiters or 'ignore' instructions for the user-supplied content before it is interpolated into the generation prompts.
Capability inventory: The skill has network access (via requests to OpenRouter) and filesystem write access (to save generated PNG and PDF files).
Sanitization: No input sanitization is performed on the raw text prompts or image data.
[EXTERNAL_DOWNLOADS]: The worker scripts (generate_slide_image_ai.py and generate_schematic_ai.py) perform network requests to openrouter.ai to access LLM services. This is a functional requirement for the skill's primary purpose and targets a well-known service provider. API credentials are appropriately managed via environment variables or local .env files, following secret management best practices.

scientific-slides