The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is subject to indirect prompt injection risks as it processes untrusted user-supplied raster images. These images could contain adversarial content or metadata designed to influence the agent's behavior during the analysis phase. \n
Ingestion points: original.png (provided by the user). \n
Boundary markers: The instructions do not specify boundary markers or 'ignore' instructions for the image analysis phase. \n
Capability inventory: The agent has the capability to write files, create directories, and execute shell commands (magick, rsvg-convert, xmllint, python3). \n
Sanitization: No sanitization is performed on the input image or its metadata. \n- [COMMAND_EXECUTION]: The workflow relies on executing local command-line tools to process images. The scripts/crop-tool.py utility uses the subprocess module to invoke ImageMagick commands. While these commands are standard for image vectorization tasks, they represent a local execution surface. \n
Evidence: scripts/crop-tool.py uses subprocess.check_output and subprocess.run to interact with the magick binary. \n
Evidence: SKILL.md and workflow/workflow-verification.md guide the agent to run rsvg-convert, xmllint, and svgo via the shell. \n- [EXTERNAL_DOWNLOADS]: The skill setup phase includes instructions for the agent to install external dependencies from well-known registries. \n
Evidence: workflow/workflow-dependencies.md specifies installing pyyaml and vtracer via pip, and optionally svgo via npm to enhance vectorization and optimization capabilities.

image-to-svg