canva-translate-design
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests text elements from external Canva designs, which introduces an attack surface where malicious instructions could be embedded in the design data.
- Ingestion points: Text elements are retrieved from designs via the
Canva:start-editing-transactiontool in the workflow described inSKILL.md. - Boundary markers: The instructions do not explicitly provide boundary markers or instructions to the model to ignore potential commands within the design text during translation.
- Capability inventory: The skill has write capabilities to the user's Canva account, including creating copies (
Canva:resize-design), performing edits (Canva:perform-editing-operations), and committing changes (Canva:commit-editing-transaction). - Sanitization: No explicit sanitization of the retrieved design text is mentioned before it is processed by the language model for translation.
Audit Metadata