The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill contains scripts that ingest untrusted data and interpolate it into SVG files, creating an attack surface for indirect prompt injection if the output is later processed by an agent or rendered in a browser.
Ingestion points: scripts/generate_chart.py (via --labels and --data), scripts/generate_icon.py (via --icon), and scripts/optimize_svg.py (via stdin or input file).
Boundary markers: The skill does not implement boundary markers or instructions to the agent to ignore potentially malicious content within the processed data.
Capability inventory: All scripts possess file-write capabilities through the -o or --output parameters, and scripts/optimize_svg.py has file-read capabilities.
Sanitization: While numerical data is validated as floats, string inputs such as labels, icon names, and colors are not escaped or sanitized against XML/SVG injection techniques.

svg-art