canvas-design
Pass
Audited by Gen Agent Trust Hub on Jun 30, 2026
Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: The
FINAL STEPsection utilizes a simulated state injection technique by claiming 'The user ALREADY said...' to force the agent into a secondary refinement loop. This attempts to bypass the agent's perception of the current user's actual feedback and override its standard processing state. - [PROMPT_INJECTION]: The skill uses high-priority markers ('CRITICAL', 'IMPORTANT', 'NON-NEGOTIABLE') and repetitive instructions to enforce a rigid 'master craftsman' persona and aesthetic style, which restricts the agent's natural instructional boundaries.
- [EXTERNAL_DOWNLOADS]: The instructions explicitly tell the agent to 'Download and use whatever fonts are needed to make this a reality,' encouraging the fetching of external assets from unspecified remote sources.
- [SAFE]: The skill includes 26 SIL Open Font License (OFL) files in the
canvas-fonts/directory to ensure the legal use of font assets, demonstrating adherence to intellectual property best practices.
Audit Metadata