Home Assistant

Home Assistant Control

This skill enables Claude to interact with a Home Assistant instance via its REST API — querying device states, controlling smart home devices, reviewing history, and working with automations.

Slash Command

When invoked via /home-assistant, ensure authentication is set up (see below), then ask the user what they'd like to do — for example: check device status, control lights, adjust the thermostat, view history, or manage automations.

Setup & Authentication

Configuration lives in ~/.config/home-assistant/.env. On first use, check if this file exists:

cat ~/.config/home-assistant/.env 2>/dev/null || echo "NOT_FOUND"

If the file exists, source it to load HA_URL and HA_TOKEN:

source ~/.config/home-assistant/.env

If the file doesn't exist, walk the user through setup:

1. Explain what's needed: a Home Assistant URL and a Long-Lived Access Token
2. Ask for their HA URL (e.g., http://192.168.1.100:8123 or https://home.example.com)
3. Guide them to create a token: Go to their HA instance → Profile (bottom-left) → Security tab → scroll to "Long-Lived Access Tokens" → click "Create Token" → give it a name like "Claude" → copy the token
4. Create the config:

mkdir -p ~/.config/home-assistant
cat > ~/.config/home-assistant/.env << 'EOF'
HA_URL="http://your-ha-address:8123"
HA_TOKEN="your_token_here"
EOF
chmod 600 ~/.config/home-assistant/.env

Replace the placeholder values with what the user provides. The chmod 600 ensures only the user can read the token.

After sourcing, verify connectivity:

source ~/.config/home-assistant/.env
curl -s -H "Authorization: Bearer $HA_TOKEN" "$HA_URL/api/"

Expected response: {"message": "API running."}. If it fails, check the URL and token with the user.

Workflow

Follow this general approach for any Home Assistant task:

1. Discover what's available

Before trying to control or query specific devices, find out what entities exist. Run the discovery script to get a formatted overview:

bash <skill-path>/scripts/discover_entities.sh

This lists all entities grouped by domain (lights, switches, sensors, etc.) with their current states. Use this output to map the user's natural language ("the kitchen light") to actual entity IDs (light.kitchen_ceiling).

If the user asks about a specific type of device, you can filter:

bash <skill-path>/scripts/discover_entities.sh light
bash <skill-path>/scripts/discover_entities.sh sensor
bash <skill-path>/scripts/discover_entities.sh climate

2. Query state

To check the current state of a specific entity:

curl -s -H "Authorization: Bearer $HA_TOKEN" \
  -H "Content-Type: application/json" \
  "$HA_URL/api/states/<entity_id>"

The response includes state (the primary value) and attributes (detailed properties like brightness, temperature, friendly name, etc.).

3. Take action (with confirmation)

Before performing any action that changes device state, describe what you're about to do and ask the user to confirm. This applies to all service calls — turning things on/off, locking/unlocking, adjusting temperatures, triggering scenes, etc. The reason: smart home actions affect the physical world and can't be undone with ctrl-z.

To call a service:

curl -s -X POST \
  -H "Authorization: Bearer $HA_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"entity_id": "<entity_id>"}' \
  "$HA_URL/api/services/<domain>/<service>"

Common examples:

• Turn on a light: domain=light, service=turn_on, data can include brightness (0-255), color_temp, rgb_color
• Turn off a light: domain=light, service=turn_off
• Toggle a switch: domain=switch, service=toggle
• Set thermostat: domain=climate, service=set_temperature, data includes temperature
• Lock/unlock: domain=lock, service=lock or unlock
• Trigger a scene: domain=scene, service=turn_on
• Run a script: domain=script, service=turn_on (or the script's entity name)

4. Review history

To see what happened with an entity over time:

curl -s -H "Authorization: Bearer $HA_TOKEN" \
  "$HA_URL/api/history/period/<timestamp>?filter_entity_id=<entity_id>&end_time=<end_timestamp>&minimal_response"

Timestamps use ISO 8601 format (e.g., 2025-01-15T08:00:00+00:00). The minimal_response flag reduces payload size.

For the logbook (human-readable event log):

curl -s -H "Authorization: Bearer $HA_TOKEN" \
  "$HA_URL/api/logbook/<timestamp>?entity=<entity_id>"

Working with automations

Home Assistant automations are entities in the automation domain. You can:

• List automations: Filter discovery output or query /api/states for entities starting with automation.
• Trigger an automation: Call automation/trigger service
• Enable/disable: Call automation/turn_on or automation/turn_off
• View automation config: The entity's attributes contain the automation's configuration

To create or modify automations, you'd need to edit the HA configuration files directly (typically automations.yaml), which is outside the scope of the REST API. If the user asks for this, let them know and offer to help draft the YAML instead.

Rendering templates

Home Assistant supports Jinja2 templates for dynamic values. To evaluate a template:

curl -s -X POST \
  -H "Authorization: Bearer $HA_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"template": "{{ states(\"sensor.temperature\") }}"}' \
  "$HA_URL/api/template"

This is useful when the user wants computed values or complex state queries.

Error handling

• 401 Unauthorized: Token is invalid or expired. Ask the user to regenerate it.
• 404 Not Found: Entity doesn't exist. Run discovery to find the correct entity_id.
• 400 Bad Request: Usually means malformed service data. Check the service's expected parameters by querying /api/services.

When a call fails, show the user the error response and suggest what to fix. Don't silently retry.

Tips for natural language mapping

Users will say things like "turn off the bedroom light" rather than giving you entity IDs. To handle this:

1. Run discovery to get all entities
2. Look at friendly_name attributes to match natural language to entity IDs
3. If ambiguous (multiple "bedroom" lights), ask the user which one they mean
4. Cache the entity list mentally within the conversation — no need to re-discover every time

Reference

For the complete REST API endpoint reference (all endpoints, parameters, and response formats), read references/rest-api.md.