The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs file processing by executing shell commands such as magick, vtracer, and npx svgo. These commands interpolate the INPUT_PNG and OUTPUT_SVG variables, which are expected to be supplied via user input.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input in the form of file paths and image data, creating a potential attack surface.
Ingestion points: The INPUT_PNG and OUTPUT_SVG variables in SKILL.md are used directly in shell command templates.
Boundary markers: The instructions do not define boundary markers or explicit warnings to ignore instructions embedded within file paths or image metadata.
Capability inventory: The skill uses subprocess calls to execute image processing tools and file system operations (rm).
Sanitization: No input sanitization or validation logic is specified within the skill's instructions.
[EXTERNAL_DOWNLOADS]: The documentation includes instructions for installing required dependencies using well-known package managers, including Homebrew for imagemagick, Cargo for vtracer, and npm for svgo.

lov-png2svg