Manual test script
For each domain action, verify:
- Golden path — speak an unambiguous command → tool called with correct args → state mutated → assistant confirms briefly in voice.
- Ambiguous reference — assistant asks for clarification, no mutation.
- Unrelated chit-chat — assistant chats briefly, no tool call.
- Empty state read-back — assistant says something meaningful, not silence.
- Reload — UI shows the last confirmed state.
- Invalid API key — session start fails cleanly, error shown in log.
- Mic denied — session start fails cleanly, error shown in log.
- Interrupt — user talks over the assistant → server VAD interrupts the response and starts a new turn.
- MCP tool (if enabled) — voice command triggers an MCP call →
approval prompt surfaced (for
require_approval: "always") → executed → confirmed. See MCP Support.