interface-craft
Pass
Audited by Gen Agent Trust Hub on Oct 19, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from multiple sources to provide design feedback and refactor animations. It functions by reading local source files, analyzing user-provided screenshots, and fetching external website content via tools like WebFetch. There are no explicit instructions or boundary markers provided to the agent to ensure it disregards natural language instructions that might be embedded within the code, images, or web pages being critiqued. This enables a potential attack vector where malicious data could influence the agent's output or subsequent actions.
- Ingestion points: The skill routes requests to read file paths (SKILL.md), evaluates pasted images or screenshots (design-critique.md), and retrieves page content from live URLs using WebFetch (design-critique.md).
- Boundary markers: Absent. The instructions do not define delimiters or provide warnings to the agent to distinguish between the UI data being analyzed and potential instructions contained within that data.
- Capability inventory: The skill utilizes file system reading, network fetching (WebFetch), and generates complex React, Solid, Svelte, and Vue code snippets for the user to implement.
- Sanitization: Absent. No logic is defined for filtering or escaping content retrieved from external files or web pages before it is processed by the agent.
Audit Metadata