canvas-design
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill instructions utilize highly authoritative and coercive language (e.g., "CRITICAL," "NON-NEGOTIABLE," "VERY IMPORTANT") to override the agent's default behavior and enforce a specific persona. It specifically injects a simulated user complaint ("The user ALREADY said 'It isn't perfect enough...'") into the agent's context. This is a technique designed to force the model into an iterative refinement loop and bypass its standard evaluation of user satisfaction.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and interpret arbitrary user input to deduce "subtle conceptual threads" and "visual philosophies."
- Ingestion points: User-provided design prompts are processed in
SKILL.mdunder the "DESIGN PHILOSOPHY CREATION" and "DEDUCING THE SUBTLE REFERENCE" sections. - Boundary markers: The skill lacks specific delimiters or "ignore embedded instructions" warnings to isolate user-provided content from the operational logic.
- Capability inventory: The agent has the capability to write files (
.md,.pdf,.png) and search local directories (./canvas-fonts). - Sanitization: There is no evidence of sanitization or validation for the conceptual DNA woven into the design from user input, which could allow malicious instructions to influence the agent's creative process.
Audit Metadata