The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied image files (PNG/JPEG) through a multi-step pipeline involving imagemagick, potrace, and python3. While the goal is image conversion, processing untrusted binary files through external tools creates a surface for indirect injection if the tools themselves or the scripts parsing their output are vulnerable to specially crafted metadata or file structures.
Ingestion points: Ingests user-provided image files (input.png) via the scripts/png2svg.sh script.
Boundary markers: None. The script directly processes the file content using shell and Python commands.
Capability inventory: The skill has significant capabilities including file read/write, shell execution (magick, potrace), and dynamic execution of Python snippets to manipulate SVG content.
Sanitization: The script performs basic validation (checking if file exists), but does not sanitize the contents of the image files against potential parser exploits in imagemagick or potrace (e.g., ImageTragick-style vulnerabilities).
[COMMAND_EXECUTION]: The skill relies on executing external binaries (magick, potrace) and invokes python3 with dynamically generated strings in the trace_single_color and composite_svg_layers functions. While the variables being interpolated are largely controlled (colors, filenames), this represents a pattern of dynamic execution that requires caution when handling user-provided filenames.

png-to-svg