excalidraw-mcp-diagramming
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes text from a live, external Excalidraw canvas, creating a surface for indirect prompt injection if an adversarial user modifies the whiteboard content.
- Ingestion points: The
describe_scene()tool inSKILL.mdandreferences/iterative-refinement.mdretrieves text data and labels directly from the active canvas into the agent context. - Boundary markers: Absent; there are no explicit delimiters or strict system guidelines instructing the model to ignore executable instructions within canvas text fields.
- Capability inventory: The skill contains local file modification capabilities through the
export_to_imageandexport_scenetools inSKILL.md. - Sanitization: Absent; text fields from the canvas elements are consumed as-is without escaping or filtering controls.
Audit Metadata