The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes local command-line tools for SVG rasterization and favicon generation.
Evidence: references/exports.md suggests using renderers such as resvg, rsvg-convert, Inkscape, sharp, CairoSVG, and ImageMagick to process SVG files.
Mitigation: The skill includes explicit instructions to "Inspect available local tools before choosing commands" and "Do not install a renderer without permission."
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process user-provided visual references which could contain adversarial instructions embedded in data.
Ingestion points: SKILL.md (Intake section) and references/curated-abstract-marks.md allow the agent to process "source logos supplied as taste references" and other user-curated preferences.
Capability inventory: The agent has the ability to generate executable SVG markup and invoke local shell commands for file conversion.
Boundary markers: The skill instructs the agent to treat references as "optional visual vocabulary" rather than fixed templates.
Sanitization: The skill explicitly forbids the inclusion of <script> tags, external resources, or embedded raster data in the generated SVG markup to prevent XSS and data exfiltration.
[DYNAMIC_EXECUTION]: The agent generates SVG XML code dynamically based on user prompts.
Evidence: SKILL.md and references/icon-draft.md guide the creation of vector geometry code.
Mitigation: The skill mandates that geometry remain concise and prohibits unsafe SVG features like internal scripts or external resource loading, which reduces the risk of malicious code being rendered by a browser or image processor.

svg-icon-designer-acrazie