canvas-design
Warn
Audited by Gen Agent Trust Hub on Apr 12, 2026
Risk Level: MEDIUMPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: The skill's metadata falsely attributes the work to 'Anthropic', which contradicts the actual author identity. This deceptive metadata is intended to impersonate a trusted organization and can lead to misjudgment of the skill's source and safety.
- [PROMPT_INJECTION]: The instructions employ a manipulative steering technique by claiming 'The user ALREADY said' that the output must be a masterpiece. This fabrication of user history is a role-play injection designed to override the agent's standard output logic and enforce a specific aesthetic constraint.
- [PROMPT_INJECTION]: The skill ingests user input to deduce conceptual themes without using any delimiters or boundary markers. 1. Ingestion points: User input referenced in SKILL.md. 2. Boundary markers: Absent. 3. Capability inventory: Creation of .md, .pdf, and .png files, and searching the font directory. 4. Sanitization: Absent.
- [EXTERNAL_DOWNLOADS]: The skill encourages the dynamic retrieval of external assets with the instruction to 'Download and use whatever fonts are needed'. The associated font license files reference official repositories and domains of well-known organizations such as Google, Vercel, IBM, and Red Hat.
Audit Metadata