canvas-design
Pass
Audited by Gen Agent Trust Hub on Sep 24, 2026
Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The 'FINAL STEP' section uses a deceptive technique by claiming the user has already provided a specific instruction ('The user ALREADY said It isn't perfect enough...'). This is designed to override the agent's current state and force it into a recursive refinement loop regardless of the actual user interaction history.\n- [EXTERNAL_DOWNLOADS]: The skill instructs the agent to 'Download and use whatever fonts are needed', which grants permission to fetch data from arbitrary, unspecified external network sources without integrity verification.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied 'subtle input or instructions' as the foundation for creative output. It lacks boundary markers or sanitization to prevent these inputs from containing hidden instructions that could manipulate the agent's behavior during the design process.\n
- Ingestion points: User instructions or input mentioned in SKILL.md used to deduce the soul or conceptual DNA of the work.\n
- Boundary markers: Absent; there are no delimiters or instructions to ignore embedded commands within user input.\n
- Capability inventory: The agent has tools for file writing (.pdf, .png, .md) and network access for downloading assets.\n
- Sanitization: Absent; the skill does not specify any validation or filtering for user-provided creative prompts.
Audit Metadata