The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill performs runtime compilation of a provided C source file (nn_assign.c) using gcc to accelerate nearest-neighbor calculations.\n
Evidence: The _ensure_nn_binary function in scripts/pipeline.py invokes gcc to compile the source, and _nn_assign_fast executes the resulting binary via subprocess.run.\n- [COMMAND_EXECUTION]: The skill invokes several external command-line utilities to perform image processing and verification tasks.\n
Evidence:\n
rsvg-convert is used for SVG to PNG conversion in the verification protocol described in SKILL.md.\n
ImageMagick (convert or magick) is used for image smoothing in the _im_smooth function within scripts/pipeline.py.\n
The gcc compiler is invoked to build the accelerated binary for quantization.\n- [EXTERNAL_DOWNLOADS]: The skill setup requires downloading and installing several well-known image processing libraries and system packages.\n
Evidence: SKILL.md instructs the user to pip install packages including opencv-python-headless, scikit-image, scipy, and scikit-learn, and to apt-get install the librsvg2-bin package.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-supplied image data, which represents a potential attack surface for indirect prompt injection.\n
Ingestion points: User-provided raster images processed by the image_to_svg function.\n
Boundary markers: The skill instructions emphasize computational analysis over visual interpretation, providing a logical boundary that prevents the LLM from directly interpreting potential text-based instructions inside images.\n
Capability inventory: The skill can execute shell commands via subprocess, compile source code, and perform file system operations.\n
Sanitization: Input images are processed via binary vision libraries (OpenCV and ImageMagick), though no specific content filtering for malicious instructions is performed.

image-to-svg