Skill: Weekly Energy Digest

Purpose

Generate and send a weekly energy efficiency digest every Sunday at 9am. Analyze device energy
consumption from the past 7 days, identify top consumers, compare against stored baselines,
and compose a structured summary with recommendations. Deliver via
notify(channel="telegram", intent="send").

When to Use

Use this skill when:

• The weekly-energy-digest scheduled task fires (cron: 0 9 * * 0, Sundays at 09:00)
• User requests "send me the weekly energy report" or similar

Workflow

Step 1: Discover Energy Sensors

1. Call ha_list_entities(domain="sensor") to get all sensor entities.
2. Filter results for energy-related entities — look for entity IDs or friendly names containing
   energy, power, kwh, consumption, watt (case-insensitive).
3. Build a list of statistic_ids for the top-level energy meter and per-device sensors
   (e.g., ["sensor.energy_total_kwh", "sensor.hvac_energy", "sensor.water_heater_energy", "sensor.kitchen_energy"]).

Step 2: Retrieve Weekly Energy Data

Call ha_get_statistics() to get the past 7 days of energy data:

ha_get_statistics(
    statistic_ids=<energy_sensor_ids>,
    start=<7 days ago, ISO 8601, midnight>,
    end=<now, ISO 8601>,
    period="day"
)

This returns daily aggregated statistics per sensor for the device breakdown.

For aggregate consumption, request an hour-aligned 168-hour window:

ha_get_statistics(
    statistic_ids=<per_device_sensor_ids>,
    start=<168 hours before the latest UTC hour boundary>,
    end=<latest UTC hour boundary>,
    period="hour"
)

Sum only finite numeric per-hour change values. Home Assistant derives change from
cumulative-energy statistics; do not integrate mean power values. Treat a series with any
missing, non-numeric, or non-finite change as unsupported. An explicit change=0 is valid
zero consumption.

If every sensor is unsupported, notify the owner that cumulative-energy statistics are
unavailable and recommend configuring a Home Assistant energy helper for power-only sensors.
Do not compute or store a baseline.

Step 3: Retrieve Baselines from Memory

Call memory_recall(subject="energy", predicate="energy_baseline") to retrieve stored baseline
consumption patterns.

Also search for device-specific baselines:

• memory_recall(subject="hvac", predicate="energy_baseline")
• memory_recall(subject="water-heater", predicate="energy_baseline")
• Any device baselines stored previously

Step 4: Compute Top Consumers and Trends

From the weekly statistics:

1. Rank devices by total consumption (sum over the week). Identify the top 5 consumers.
2. Calculate percentage share of each device relative to total consumption.
3. Compare against baselines:
   • Is this week's total higher or lower than typical?
   • Are any devices consuming 20%+ above their baseline? (anomaly)
   • What was the peak consumption day?
4. Identify anomalies (flag if present):
   • Device consuming 2x or more than baseline (high severity)
   • Device consuming 20-50% above baseline (medium severity)
   • Unexpected always-on consumption (low but persistent)

Step 5: Generate Recommendations

Based on top consumers and anomalies, compose 2-3 actionable recommendations:

• High HVAC usage: "Consider lowering the heating setpoint by 2°F at night"
• Water heater anomaly: "Water heater ran longer than usual — check for leaks or tank issues"
• Standby waste: "TV and chargers in standby mode use ~5W continuously"

Limit to 2-3 recommendations. Prioritize by impact.

Step 6: Store Energy Patterns in Memory

After analysis, persist key findings:

memory_store_fact(
    subject="energy",
    predicate="energy_baseline",
    content="Weekly consumption: <X> kWh total. Top consumers: HVAC (<Y>%), water heater (<Z>%)",
    permanence="standard",
    importance=6.0,
    tags=["energy", "weekly-digest", "baseline"]
)

If an anomaly was detected:

memory_store_fact(
    subject=<device_name>,
    predicate="energy_spike",
    content="<device> consumed <X> kWh this week — <Y>% above baseline",
    permanence="volatile",
    importance=7.5,
    tags=["energy", "anomaly", <device_name>]
)

Step 7: Compose and Send the Digest

Format the digest as a structured message:

Weekly Energy Digest — [Date range, e.g. "Feb 17-23"]

Total: [X] kWh  ([+/-Y]% vs. typical)
Peak day: [Weekday] with [Z] kWh

Top Consumers:
1. HVAC — [X] kWh ([Y]%)
2. Water Heater — [X] kWh ([Y]%)
3. Kitchen Appliances — [X] kWh ([Y]%)
4. Washer/Dryer — [X] kWh ([Y]%)
5. Lighting — [X] kWh ([Y]%)

[Anomaly alert if present:]
  Water heater ran 6h longer than usual on Tuesday — possible tank issue.

Recommendations:
• [Recommendation 1]
• [Recommendation 2]
[• Recommendation 3 if applicable]

Savings vs. baseline: [X] kWh saved / [Y]% below average (or: usage is on target)

Send via:

notify(
    channel="telegram",
    intent="send",
    subject="Weekly Energy Digest — [Date range]",
    message=<formatted_digest>,
)

Use intent="send" — this is a scheduled proactive delivery, not a reply.

Exit Criteria

• ha_list_entities(domain="sensor") called to discover energy sensors
• ha_get_statistics() called with an hour-aligned 168-hour window and period="hour" to
  retrieve finite per-hour change values
• memory_recall() called to retrieve stored energy baselines
• Top 5 consumers ranked; anomalies identified
• 2-3 recommendations generated
• memory_store_fact() called to update energy baseline and any anomaly facts
• Digest composed and sent via notify(channel="telegram", intent="send")
• Session exits — no interactive follow-up in this session

Common Failure Modes

No Energy Sensors Found

• Alert via notify(channel="telegram", intent="send"): "Could not find energy sensors in Home Assistant. Weekly digest
  unavailable. Check that energy monitoring is configured in HA."
• Exit cleanly.

Partial Data (Sensor Gaps)

• Omit any sensor whose hourly series has a missing, non-numeric, or non-finite change.
• Name omitted sensors visibly and recommend configuring a Home Assistant energy helper.
• Report supported device consumption only. Do not estimate gaps or present a whole-home total,
  trend, savings claim, percentage share, or overall baseline from incomplete coverage.
• Do not skip the digest when at least one supported device series remains.

No Stored Baselines Yet

• This may be the first digest. Compose digest without trend comparison.
• Store today's data as the initial baseline for future comparison.
• Note in digest: "This is your first energy digest — we'll compare against this baseline next week."