canvas-design
Pass
Audited by Gen Agent Trust Hub on Mar 6, 2026
Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: The skill employs a 'state injection' technique in the 'FINAL STEP' section of
SKILL.md, where it instructs the agent that 'The user ALREADY said' certain phrases to force a refinement loop. This is a pattern used to override an agent's current state or prior instructions. - [EXTERNAL_DOWNLOADS]: The instructions in
SKILL.mdtell the agent to 'Download and use whatever fonts are needed' to complete the artistic work. This encourages the agent to interact with external network resources to fetch assets, which could potentially lead to the acquisition of untrusted content. - [PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection. Ingestion points: User-provided 'subtle input' foundations are processed in step 1. Boundary markers: There are no delimiters or 'ignore embedded instructions' warnings for the external data. Capability inventory: The agent has the ability to write multiple file types including
.md,.pdf, and.png. Sanitization: No sanitization or validation of the user foundation is specified before it is used to generate the design philosophy.
Audit Metadata