ui-from-screenshot
Pass
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied images (screenshots and mockups), which represents an attack surface where hidden text instructions could attempt to influence the agent's behavior. However, the skill implements strong procedural boundaries that mitigate this risk.
- Ingestion points: User-provided screenshots, mockups, or design exports (SKILL.md).
- Boundary markers: Present. The instructions mandate a multi-step 'Read the image' phase where the agent must explicitly document structure and measurements before writing any code, which serves as a cognitive check.
- Capability inventory: Frontend UI implementation, including component creation and layout modification.
- Sanitization: The instructions include explicit rules to ignore branding, logos, and proprietary content from the source image, focusing only on layout patterns and generic placeholders.
- [SAFE]: The skill includes security-positive instructions, such as explicitly forbidding the agent from adding external servers or accounts (MCP servers) without a direct user request and requiring user sign-in for optional features.
Audit Metadata