The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied text prompts and attached images in scripts/generate_slide_image_ai.py, which are then sent to an external AI service. This creates an attack surface where malicious instructions could be embedded in the data processed by the agent.
Ingestion points: User-provided prompt string and image attachments processed in scripts/generate_slide_image.py and scripts/generate_slide_image_ai.py.
Boundary markers: The generate_slide_image_ai.py script wraps user input within specific instructional blocks (FULL_SLIDE_GUIDELINES and VISUAL_ONLY_GUIDELINES) to guide the model's behavior.
Capability inventory: The skill can execute shell commands via subprocess (e.g., calling pdflatex in scripts/validate_presentation.py and script orchestration in scripts/generate_slide_image.py) and perform network operations via the requests library.
Sanitization: No specific sanitization or filtering is performed on the user-provided text prompts before inclusion in the AI model request.
[COMMAND_EXECUTION]: The skill uses subprocess calls to handle core tasks like LaTeX compilation and script orchestration.
Evidence: scripts/validate_presentation.py executes pdflatex to compile and validate LaTeX Beamer presentations.
Evidence: scripts/generate_slide_image.py uses subprocess.run to invoke the internal generate_slide_image_ai.py script.
[EXTERNAL_DOWNLOADS]: The skill makes network requests to an external service to perform slide generation and review.
Evidence: scripts/generate_slide_image_ai.py uses the requests library to communicate with https://openrouter.ai/api/v1 for image generation and quality assessment.

alterlab-scientific-slides