autofigure
Pass
Audited by Gen Agent Trust Hub on Oct 17, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill fetches model weights for background removal (RMBG-2.0) from HuggingFace, an established platform for hosting machine learning resources.
- [INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect instructions embedded within processed data.
- Ingestion points: Untrusted data enters the agent context through user-provided raster images (PNG/JPEG) via the
image_pathparameter in SKILL.md. - Boundary markers: Explicit delimiters for the vision-to-XML reconstruction process are not defined in the skill body, with the implementation logic residing within the specific tools.
- Capability inventory: The skill is capable of writing generated XML content to the local filesystem (
template.drawio,optimized_template.drawio,final.drawio). - Sanitization: The skill incorporates automatic XML validation and LLM-based repair iterations to ensure the output conforms to the expected DrawIO schema.
Audit Metadata