ha-event-response

name	ha-event-response
description	Home Assistant event classification and response procedure for safety, environmental drift, automation failures, and routine changes.
version	1.0.0
tools_required	["ha_call_service","ha_get_history","ha_list_entities","memory_recall","memory_store_fact","notify"]

Home HA Event Response Skill

Purpose

Load this skill when a routed Home Assistant event must be classified and handled as a proactive home state change.

HA Event Response Patterns

When the Home butler receives a routed ingestion event from the Switchboard with source.channel = "home_assistant", it processes the event as a proactive home state change notification. The butler must determine the appropriate response using its HA tools and memory.

Event Classification

Classify every HA-sourced event into one of four categories before responding:

Category	Trigger	Response
Safety-critical	Lock/cover entity transitions to unexpected state (e.g. door unlocked at night, garage opened while away)	Notify owner immediately; store volatile fact
Environmental drift	Sensor reading falls outside owner's stored comfort preferences	Take corrective action if available, else notify owner
Automation failure	Automation triggered with unexpected outcome, or entity transitions to unavailable	Store volatile device_issue fact; include in next health check
Routine acknowledgment	Expected state change (lights on during normal hours, ordinary temperature fluctuation)	Store volatile fact silently for pattern analysis; do NOT notify

Safety-Critical Events

Trigger conditions:

• lock entity state changes to unlocked during nighttime hours or while the home is set to "away" mode
• cover entity (garage door, gate, skylight) opens unexpectedly
• Security-related binary_sensor triggers (motion in vacant areas, door/window sensor while armed)

Required response:

1. Notify owner immediately: notify(channel="telegram", intent="proactive", message=<alert>)
2. Store volatile fact with importance >= 8.0
3. Do NOT wait for owner confirmation; this is a push alert

Memory fact:

memory_store_fact(
    subject="front-door-lock",          # entity-specific subject
    predicate="device_issue",
    content="Front door unlocked at 11:42pm — unexpected; owner notified",
    permanence="volatile",
    importance=9.0,
    tags=["security", "lock", "urgent", "alert"]
)

Environmental Drift Events

Trigger conditions:

• Temperature sensor reading deviates from stored comfort_preference by more than the significance threshold
• Humidity sensor reading falls outside owner's preferred range
• Air quality sensor reaches an alert threshold

Required response:

1. Check memory for applicable comfort_preference facts: memory_recall(subject=<room>, predicate="comfort_preference")
2. Look for a matching scene or automation: ha_list_entities(domain="climate") or ha_list_entities(domain="scene")
3. If corrective action is available: Execute it silently (e.g., adjust thermostat), then store a comfort_deviation fact
4. If manual intervention is needed: Notify owner with context and suggested action

Memory fact (corrective action taken):

memory_store_fact(
    subject="living-room",
    predicate="comfort_deviation",
    content="Temperature drifted to 78°F (preference: 71-73°F); thermostat auto-adjusted to 71°F",
    permanence="volatile",
    importance=5.0,
    tags=["temperature", "comfort", "auto-corrected"]
)

Automation Failure Events

Trigger conditions:

• An automation entity triggers but the expected state change does not occur
• An entity transitions to unavailable or unknown state
• A script domain entity exits with an error state

Required response:

1. Store a volatile device_issue fact with appropriate failure tags
2. Do NOT notify the owner immediately for single occurrences; include in the next scheduled device health check
3. If the entity has been unavailable for more than one hour (check ha_get_history), escalate to owner notification

Memory fact:

memory_store_fact(
    subject="basement-sensor",          # use the specific entity's subject
    predicate="device_issue",
    content="Entity transitioned to unavailable at 03:14am; first occurrence",
    permanence="volatile",
    importance=6.0,
    tags=["unavailable", "device", "maintenance", "health-check"]
)

Routine State Change Acknowledgment

Trigger conditions:

• Light state changes during expected hours (on during morning/evening, off at bedtime)
• Temperature fluctuation within normal daily range and within comfort preferences
• Expected automation triggers (scheduled scene activations, timed device changes)

Required response:

1. Store a volatile fact for pattern tracking
2. Do NOT notify the owner
3. Do NOT call any HA services (no corrective action needed)

Memory fact:

memory_store_fact(
    subject="living-room-light",
    predicate="usage_pattern",
    content="Light turned on at 6:47am — consistent with morning routine pattern",
    permanence="volatile",
    importance=2.0,
    tags=["routine", "pattern", "light"]
)

Interactive Response Mode: HA Event Handling

When an HA event arrives via a Telegram-like interactive channel (i.e., the ingest envelope includes source_channel = "telegram_bot" re-routing from a user-facing follow-up), treat the session as interactive. For purely system-generated HA events, the source_channel will be "home_assistant" and the session is non-interactive; notify() is the only output path.

Non-interactive HA events (source_channel = "home_assistant")

Use the four response categories above. Do NOT produce conversational text output; it is discarded in headless sessions. All owner-facing communication MUST go through notify().

Event category	Response mode
Safety-critical	React + Reply: emoji (🚨) plus alert message via notify(intent="proactive")
Environmental drift (auto-corrected)	Silent: store fact, no notify
Environmental drift (manual needed)	Affirm via notify(intent="proactive"): state problem and suggested action
Automation failure (first occurrence)	Silent: store fact, no notify
Automation failure (extended unavailability)	Affirm via notify(intent="proactive")
Routine acknowledgment	Silent: store fact only

Example: Safety-critical HA event (non-interactive)

Incoming event: lock.front_door → unlocked at 11:42pm

Actions:

1. memory_recall(subject="home-mode", predicate="away_status") to check if away/nighttime mode
2. memory_store_fact(subject="front-door-lock", predicate="device_issue", content="Front door unlocked at 11:42pm unexpectedly", permanence="volatile", importance=9.0, tags=["security", "lock", "urgent", "alert"])
3. notify(channel="telegram", intent="proactive", message="🚨 Front door unlocked at 11:42pm. Was this you? Reply ✅ to confirm or I can re-lock it for you.")

Example: Environmental drift with corrective action (non-interactive)

Incoming event: sensor.living_room_temperature → 78°F (comfort preference: 71-73°F)

Actions:

1. memory_recall(subject="living-room", predicate="comfort_preference") to get comfort range
2. ha_list_entities(domain="climate", area="living_room") to find thermostat
3. ha_call_service(domain="climate", service="set_temperature", target={"area_id": "living_room"}, data={"temperature": 72})
4. memory_store_fact(subject="living-room", predicate="comfort_deviation", content="Temperature drifted to 78°F; auto-adjusted thermostat to 72°F", permanence="volatile", importance=5.0, tags=["temperature", "comfort", "auto-corrected"])
5. (No notify; corrective action taken silently)

Example: Automation failure / entity unavailable (non-interactive)

Incoming event: sensor.basement_motion → unavailable

Actions:

1. ha_get_history(entity_ids=["sensor.basement_motion"], start="<2 hours ago ISO>", end="<now ISO>") to check how long it's been unavailable
2. If unavailable < 60 min: memory_store_fact(subject="basement-motion", predicate="device_issue", content="Entity unavailable since 02:30am; monitoring", permanence="volatile", importance=6.0, tags=["unavailable", "device", "maintenance", "health-check"])
3. If unavailable >= 60 min: same store + notify(channel="telegram", intent="proactive", message="Basement motion sensor has been offline for over an hour. It may need attention.")

Example: Routine acknowledgment (non-interactive, silent)

Incoming event: light.kitchen → on at 7:02am

Actions:

1. memory_store_fact(subject="kitchen-light", predicate="usage_pattern", content="Light turned on at 7:02am — consistent with morning routine", permanence="volatile", importance=2.0, tags=["routine", "pattern", "light"])
2. (No notify, no service call)