The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the subprocess.run function across several scripts (scripts/generate_slide_image.py, scripts/generate_schematic.py, and scripts/validate_presentation.py) to execute internal Python sub-scripts and system utilities like pdflatex. These executions use list-based arguments which mitigate shell injection risks and are necessary for the presentation generation and validation workflows.\n- [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because untrusted user input is directly interpolated into instructions sent to the external LLM services in scripts/generate_slide_image_ai.py and scripts/generate_schematic_ai.py.\n
Ingestion points: User-provided slide and schematic descriptions are ingested via command-line arguments (prompt) and passed to AI generation functions.\n
Boundary markers: The system prompts for Nano Banana 2 and Gemini 3.7 Flash do not employ delimiters (such as XML tags or unique markers) to separate user content from instruction blocks.\n
Capability inventory: The agent possesses the ability to execute local commands via subprocess.run, perform network requests to openrouter.ai, and modify the local filesystem.\n
Sanitization: There is no evidence of filtering, escaping, or validation of the user input before it is concatenated into the final prompt string.

scientific-slides