The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run in scripts/render_svg_icon_png.py to execute the rsvg-convert utility.
The command is executed using a list of arguments rather than a shell string, which mitigates common command injection vulnerabilities.
The inputs, such as filenames and sizes, are validated or derived from internal logic before being passed to the command.
[INDIRECT_PROMPT_INJECTION]: The skill represents an attack surface for indirect prompt injection by ingesting SVG files from external sources (svgrepo.com, svgfind.com) and passing them to a vision-capable LLM for evaluation.
Ingestion points: Files are downloaded via scripts/acquire_svgrepo_svg.py and manual processes described in references/svgfind_acquisition.md.
Boundary markers: The vision prompt template in references/icon_review_rubric.md provides structure but lacks explicit instructions for the model to ignore potential instructions embedded within SVG metadata or XML comments.
Capability inventory: The skill can perform network requests to fetch icons and execute local rendering commands via scripts/render_svg_icon_png.py.
Sanitization: scripts/acquire_svgrepo_svg.py performs basic checks to ensure the payload is an SVG and not a security challenge page, but it does not sanitize the XML content for potentially malicious instructions targetting the vision model.

select-svg-icon