The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill accepts natural language descriptions from users and interpolates them into prompt templates sent to external AI models. \n
Ingestion points: User-provided diagram descriptions are passed as the prompt argument to scripts/generate_schematic.py, which flows to the AI generation logic in generate_schematic_ai.py. \n
Boundary markers: The prompt templates use simple markers like USER REQUEST: but do not implement robust sanitization or escaping of the user input. \n
Capability inventory: The skill has the capability to execute shell commands (pdflatex, convert, open) via the compilation scripts. \n
Sanitization: No specific input validation or filtering is performed on the user prompt before interpolation. \n- [COMMAND_EXECUTION]: The skill uses subprocess.run to call system utilities for compiling LaTeX and processing images. \n
Evidence: scripts/compile_tikz.py executes pdflatex to compile TikZ code, convert (ImageMagick) to generate PNGs, and system tools like open or xdg-open to preview results. \n- [EXTERNAL_DOWNLOADS]: The skill communicates with an external service to perform its primary function. \n
Evidence: scripts/generate_schematic_ai.py makes HTTPS requests to openrouter.ai to access the google/gemini-3-pro-image-preview model for iterative diagram generation and quality review.

scientific-schematics