openai-frontend-design
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill's core workflow involves processing untrusted external data (user requirements and AI-generated design concepts) to generate code, which presents a surface for indirect prompt injection.\n
- Ingestion points: User-provided project briefs and image-generated design concepts are ingested and parsed to define the application structure, content, and component logic (SKILL.md, references/imagegen-website-concepts.md).\n
- Boundary markers: The instructions lack explicit boundary markers or delimiters to isolate untrusted input data from the agent's operational logic during the code synthesis phase (SKILL.md).\n
- Capability inventory: The agent possesses file-writing capabilities (React/Vite project files) and the ability to execute browser automation tools such as Playwright and IAB for verification (SKILL.md).\n
- Sanitization: No specific sanitization or validation logic is defined to filter potentially malicious instructions embedded within the design concepts or user briefs before they are used in code generation.\n- [DYNAMIC_EXECUTION]: The skill involves the generation and subsequent execution of web application code as a primary function.\n
- Evidence: The agent is instructed to implement design systems using React and Vite and then execute the resulting application in a browser environment to perform visual fidelity checks (SKILL.md, Implementation and Verification sections).
Audit Metadata