The Agent Skills Directory

[DATA_EXFILTRATION]: The script generate_slide_image_ai.py performs network requests to openrouter.ai to generate and review slide images. These requests send user-provided prompts and attached images to an external API. This is the intended core functionality of the skill and targets a well-known service.
[COMMAND_EXECUTION]: Several scripts utilize subprocess.run to perform legitimate tasks. For example, generate_slide_image.py executes an internal AI generation script, and validate_presentation.py calls pdflatex to verify LaTeX Beamer compilations. These executions use list-based arguments, which mitigates the risk of shell injection.
[EXTERNAL_DOWNLOADS]: The skill requires several standard Python libraries for handling PDFs, images, and PowerPoint files, including requests, Pillow, pymupdf, PyPDF2, and python-pptx.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it is designed to ingest research data (via the research-lookup skill) and user-provided figures to generate AI prompts without robust sanitization or boundary delimiters.
Ingestion points: Research results from the research-lookup skill and user-provided image files are processed in SKILL.md and generate_slide_image_ai.py.
Boundary markers: The prompt templates in generate_slide_image_ai.py lack strict delimiters (e.g., XML tags or clear 'ignore instructions' warnings) when interpolating user content.
Capability inventory: The skill has capabilities to write files, execute shell commands (Bash), and perform network operations via requests.
Sanitization: There is no evidence of input validation or escaping for the strings interpolated into the AI prompts.

scientific-slides