gpt-taste
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill uses authoritative directives such as 'CORE DIRECTIVE' and 'BANNED' to override the model's default stylistic behaviors and enforce specific design patterns. This is intended to improve design quality but employs strong persona-enforcement language to break statistical defaults.
- [EXTERNAL_DOWNLOADS]: The instructions require the agent to use the 'picsum.photos' service for generating image placeholders in the UI code. This is a well-known service for design prototyping and placeholder assets.
- [INDIRECT_PROMPT_INJECTION]: The skill processes user prompts to generate a deterministic seed for UI layout variation. This creates an ingestion surface where external input influences the generation of React and GSAP code. 1. Ingestion points: User input prompt (SKILL.md). 2. Boundary markers: Not specified for user input data. 3. Capability inventory: Generates React, Tailwind CSS, and GSAP code (SKILL.md). 4. Sanitization: No explicit sanitization or escaping of prompt content is described before interpolation into the design process.
Audit Metadata