ln-13-interaction-design-builder

Pass

Audited by Gen Agent Trust Hub on Oct 18, 2026

Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The 'Execution contract' contains instructions that attempt to override the agent's internal reasoning and safety posture. Specifically, the instruction 'do not invent approval gates from caution' encourages the agent to bypass precautionary measures unless a specific instruction source is cited. Additionally, the rigid verdict system ('apply the skill's verdict for the remaining gap instead of waiting or guessing') attempts to force the agent to reach a conclusion even when it might otherwise need to seek user clarification or halt for safety.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data through web browsing and UI inspections, creating a surface for indirect injection attacks.
  • Ingestion points: The skill ingests data via the Browser, screenshots, accessibility inspection, and User-supplied scenarios (SKILL.md).
  • Boundary markers: The instructions do not provide delimiters or warnings to the agent to distinguish between its internal instructions and the content it observes in external UIs or requirements documents.
  • Capability inventory: The skill utilizes browser access, filesystem read/write (for design artifacts), and screenshotting tools.
  • Sanitization: There is no evidence of sanitization or validation of external content before the agent interprets it to design flows and states.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 18, 2026, 05:51 PM