raster-to-clean-svg
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted raster images and agent-generated SVG files, creating a vulnerability surface for file-based attacks. Evidence: 1. Ingestion points: reference.png and agent-generated candidate.svg are processed by scripts probe_raster.py and validate_svg.py. 2. Boundary markers: Structural preflight checks are performed on SVG input. 3. Capability inventory: Uses CairoSVG or resvg_py to render SVGs and writes diagnostic output to the local filesystem. 4. Sanitization: validate_svg.py implements checks to reject forbidden elements like scripts and external resource references.
- [SAFE]: The skill uses only standard, well-known Python packages for image processing and includes comprehensive security validation to ensure the safety of its outputs.
Audit Metadata