canvas-design
Pass
Audited by Gen Agent Trust Hub on Sep 10, 2026
Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill uses a context-manipulation technique to force the agent into a specific behavior state.
- Evidence: In
SKILL.md, the section 'FINAL STEP' includes the instruction: 'IMPORTANT: The user ALREADY said "It isn't perfect enough. It must be pristine..."'. This attempts to simulate a previous user interaction to bypass the agent's standard output thresholds. - [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied instructions to generate visual art and design philosophies, creating a potential vector for indirect injection.
- Ingestion points: User-provided creative prompts and subtle input foundations (specified in
SKILL.md). - Boundary markers: None identified; the skill encourages the agent to use user input as a 'foundation' for creative freedom.
- Capability inventory: The skill creates and writes local files including
.md,.pdf, and.png(specified inSKILL.md). - Sanitization: No explicit sanitization or input validation logic is present to handle potentially malicious instructions embedded in user-provided design descriptions.
Audit Metadata