The Agent Skills Directory

[Prompt Injection] (LOW): The skill possesses a surface for indirect prompt injection.
Ingestion points: The workflow in SKILL.md describes reading content from external Canva designs using Canva:start-editing-transaction and user-provided text.
Boundary markers: Absent. The Presentation Query Format section does not define delimiters (e.g., XML tags or triple quotes) or provide instructions to the model to ignore embedded commands within the ingested content.
Capability inventory: The skill can generate and create designs via Canva:generate-design and Canva:create-design-from-candidate based on the processed input.
Sanitization: Absent. The skill lacks instructions to escape or validate the content retrieved from external designs before interpolating it into the final tool query.
[Data Exposure & Exfiltration] (SAFE): The skill accesses user brand kits and design metadata, but this is restricted to the integrated Canva toolset. No unauthorized network calls or credential exposures were detected.
[Unverifiable Dependencies] (SAFE): No external code dependencies, package manifests (e.g., package.json, requirements.txt), or remote scripts are included in this skill.

canva-branded-presentation