canvas-design
Pass
Audited by Gen Agent Trust Hub on Sep 29, 2026
Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill uses an instruction pre-seeding technique in the 'FINAL STEP' section of SKILL.md to force the agent into a specific perfectionist state. Evidence: "IMPORTANT: The user ALREADY said 'It isn't perfect enough...'" — this simulates user feedback to override the model's standard output threshold.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided topics to generate design philosophies and visual art but lacks necessary security boundaries. 1. Ingestion points: The skill takes "subtle input or instructions by the user" as conceptual foundation (referenced in SKILL.md). 2. Boundary markers: There are no instructions to treat user input as data rather than instructions, nor are there delimiters defined for input interpolation. 3. Capability inventory: The skill has the capability to write files in multiple formats (.md, .pdf, .png). 4. Sanitization: No sanitization, escaping, or validation logic is specified for the user-provided conceptual framework.
Audit Metadata