vui-integrationlisted

# VUI Integration Build voice as a **layer over an existing UI**, never a replacement for it. This skill encodes hard-won research and a shipped reference implementation: the multimodal division of labor, the six voice components, conversation and motion rules, a closed intent schema pattern, failure-first design, and the trust model that makes voice acceptable in serious enterprise contexts. ## The Non-Negotiable Principles Apply these to every decision. When a user request conflicts with one, explain the trade-off before proceeding. 1. **Voice for intent, screen for information density.** Voice is the fastest way to ask; the screen is the only sane way to show many rows. Never read a table or list aloud. Read back the count and the exception, then stop. Every spoken response must pass the one-breath test (fits in one spoken breath, 2-3 short sentences max). 2. **One source of truth.** Voice and mouse write to the same state. Every spoken number must be computed from the same data and predicates as the on-screen components. The read-back can never disagree with the screen; the day it says fourteen while the screen shows twelve, user trust is gone permanently. Structure the code so this is impossible, not merely tested. 3. **Multimodal parity.** Every voice action has a visible, clickable equivalent, and every spoken output has an on-screen text equivalent. The existing GUI remains 100% functional with voice unavailable (unsupported browser, denied mic, noisy room). Voice