gpt-taste
Pass
Audited by Gen Agent Trust Hub on Aug 26, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill employs strong, imperative language and overrides such as "CORE DIRECTIVE", "FORBIDDEN", and "BANNED FOREVER" to suppress default model behaviors and enforce a specific award-winning design engineer persona.
- [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to process user prompts to generate a "deterministic seed" (using prompt character counts) to drive its layout randomization logic. This creates a predictable surface where external input influences the structural execution of the task.
- Ingestion points: User prompt character count is used in the simulated randomization logic in SKILL.md.
- Boundary markers: Absent; there are no instructions to sanitize the user prompt or separate it from the logic-driving character count calculation.
- Capability inventory: The skill has the capability to generate and suggest the installation of UI libraries and complex React code.
- Sanitization: None; the skill relies on raw user input metrics for its core "Python-driven" simulation.
Audit Metadata