taste-design
Pass
Audited by Gen Agent Trust Hub on Jun 17, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted user input to generate design instructions that are later interpreted by other AI agents.
- Ingestion points: The skill processes a user-provided 'vibe description' in the SKILL.md instructions to determine design parameters like density, variance, and motion.
- Boundary markers: The instructions do not specify any delimiters or 'ignore' warnings to protect against malicious instructions embedded within the user's style description.
- Capability inventory: The skill utilizes the 'Write' tool to generate the DESIGN.md file, which serves as a source of truth for other agents (e.g., Google Stitch, Antigravity).
- Sanitization: There is no evidence of input validation or sanitization for the user-provided style description before it is interpolated into the generated output.
Audit Metadata