skills/yodaos-project/aiui/aiui-dev/Gen Agent Trust Hub

aiui-dev

Pass

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [SAFE]: The skill provides comprehensive technical documentation for the AIUI framework. It correctly outlines the required project structure, registration patterns, and lifecycle methods (documented in references/framework.md and references/events.md) needed to build functional agents.
  • [INDIRECT_PROMPT_INJECTION]: The documentation for the a2ui component (in references/components/a2ui.md) describes a surface for indirect prompt injection where UI is rendered directly from AI-generated JSON commands.
  • Ingestion points: The commands attribute of the <a2ui> component and the context.write() method which accept raw string output from a language model.
  • Boundary markers: The current documentation does not explicitly emphasize the use of boundary markers or instructions to ignore embedded UI commands within the AI response.
  • Capability inventory: The framework provides agents with capabilities including network access (fetch, wx.request), local storage (wx.setStorage), and device sensor access (Bluetooth, Geolocation).
  • Sanitization: There is no documented requirement for sanitizing or validating the JSON instructions before they are processed by the a2ui component in the provided examples.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 15, 2026, 05:14 AM
Security Audit — agent-trust-hub — aiui-dev