canvas-design
Pass
Audited by Gen Agent Trust Hub on Aug 18, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill instructions attempt to override agent behavior by falsifying the user's conversation history: 'IMPORTANT: The user ALREADY said "It isn't perfect enough. It must be pristine, a masterpiece if craftsmanship..."'. This is an adversarial technique designed to force the model into a specific state or persona by presenting fabricated prior user intent.
- [PROMPT_INJECTION]: The skill uses strong, coercive language ('CRITICAL', 'NON-NEGOTIABLE', 'STOP and instead ask') to bypass standard model reasoning and enforce specific high-level output constraints.
- [PROMPT_INJECTION]: The skill processes untrusted user data to generate intermediate instructions (the 'Design Philosophy') without implementing boundary markers or sanitization. This creates an indirect prompt injection surface where a user could potentially supply instructions that influence the subsequent file creation process.
- Ingestion points: User input enters the context during the 'DESIGN PHILOSOPHY CREATION' phase in SKILL.md.
- Boundary markers: None provided; the skill lacks delimiters or instructions for the agent to ignore instructions embedded within user input.
- Capability inventory: The skill has the capability to write to the file system (.md, .pdf, .png files) based on these generated instructions.
- Sanitization: No validation or escaping is applied to user-supplied text before it is used to build the design philosophy.
Audit Metadata