Add Home Assistant multi-agent system demo using Google ADK

[Copilot]

Multi-agent home automation demo in google-agent-kit/home_assistant/ using Google ADK's sub-agent orchestration pattern. All external calls are mocked -- no API keys or real integrations required.

Architecture

• Orchestrator (agents/orchestrator_agent.py) -- routes by intent keywords to specialist sub-agents via ADK's sub_agents pattern
• Devices Agent -- get_device_state / set_device_state for lights, locks, thermostat
• Schedule Agent -- get_calendar_events / create_calendar_event
• Info Agent -- get_weather / get_shopping_list / add_to_shopping_list

Components

• tools/ -- 7 deterministic mock tools returning hardcoded dicts. set_device_state persists to in-memory DB for read-after-write consistency.
• state/store.py -- module-level dict shared across agents, pre-populated with household defaults (who_is_home, preferred_movie_brightness, etc.)
• agent.py -- exports root_agent (the orchestrator) for ADK discovery
• main.py -- async entry point using InMemoryRunner with interactive console mode

Tests

36 unit tests covering state store CRUD, all mock tool return shapes + persistence, and agent wiring (tool bindings, sub-agent registration, routing keywords in instructions).

# Agent wiring is verified without LLM calls
def test_has_all_sub_agents(self):
    sub_names = {a.name for a in orchestrator_agent.sub_agents}
    assert sub_names == {"devices_agent", "schedule_agent", "info_agent"}

def test_does_not_have_tools(self):
    assert orchestrator_agent.tools == []

---

📍 Connect Copilot coding agent with Jira, Azure Boards or Linear to delegate work to Copilot in one click without leaving your project management tool.

[danielstegeman]

@copilot review this pr

[Copilot]

@copilot review this pr

Code Review Summary

This PR implements a Home Assistant multi-agent system using Google ADK. Here are the key findings:

Architecture (Good):

• Clean separation: orchestrator delegates to 3 specialist agents (devices, schedule, info)
• Each specialist owns its domain tools -- orchestrator has no tools, only sub-agents
• 7 mock tools with deterministic returns -- no real API calls
• State store properly isolates mutable state with reset_state() for test isolation

Tests (36 passing):

• State store: CRUD operations, defaults, isolation
• Tools: return shapes, persistence, edge cases
• Agents: wiring verification (tools, sub-agents, routing keywords)

Potential improvements for production:

1. _DEVICE_DB in device_tools.py is module-level mutable state -- consider moving to the shared state store for consistency
2. get_shopping_list() initializes default items on first call -- could race with reset_state() in concurrent scenarios
3. Type hints could be stricter (e.g., Any -> specific types in state store)

Security: CodeQL clean, no dependency advisories for google-adk 1.26.0

Overall, the implementation follows ADK patterns well and meets the spec requirements.