design-spatial

Pass

Audited by Gen Agent Trust Hub on Jul 31, 2026

Risk Level: SAFEREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructions recommend using python3 -m http.server or npx serve to host static web content for visual rendering and evaluation.
  • [REMOTE_CODE_EXECUTION]: The skill provides scripts/layout-audit.js, a diagnostic utility intended to be executed in a browser environment via browser_evaluate to measure layout metrics such as collision, contrast, and alignment.
  • [EXTERNAL_DOWNLOADS]: Refers to design and optical craft guidelines hosted in Vercel Labs' official GitHub repository.
  • [PROMPT_INJECTION]: The skill defines mandatory "BLOCKING GATES" (e.g., horizontal overflow checks) that the agent must satisfy before considering a task complete, which acts as a process-oriented constraint.
  • [PROMPT_INJECTION]: The skill identifies a surface for Indirect Prompt Injection by instructing the agent to critique rendered UI content:
  • Ingestion points: Rendered web UI content processed via browser_take_screenshot and DOM data analyzed via browser_evaluate (scripts/layout-audit.js).
  • Boundary markers: Absent; the skill does not specify delimiters or protective instructions for the external UI content being evaluated.
  • Capability inventory: The agent utilizes shell commands (python3, npx) and browser control tools (browser_evaluate, browser_take_screenshot).
  • Sanitization: Absent; the skill performs direct visual and geometric analysis of rendered output without pre-processing or filtering.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 31, 2026, 04:35 PM
Security Audit — agent-trust-hub — design-spatial