just say it
WoowTech Prompt Library

Not sure what to ask? Start with these 40 prompts

The hardest part of using Pi Agent is often knowing what to ask. These 40 copy-ready prompts cover real household tasks. Every prompt requests an English response; replace the example rooms, devices, schedules, services, and thresholds with details from your home.

What this library is for

These examples show how to state a requirement clearly: describe the goal, scope, exceptions, and approval boundary. You do not need to memorize commands or special syntax.

Select Copy on a prompt card, paste the text into Pi Agent, and replace the examples with your own details. If you have not started a Session yet, see Chapter 7: Start your first conversation.

Concept: A prompt is not a magic formula. It simply shows which details matter. Adapt each example to your devices, and review proposed actions before applying them.

How to use these prompts

  1. Choose the closest example

    Select Copy, then paste the prompt into the Pi Agent composer.

  2. Replace the example details

    Use your actual room names, devices, times, routes, and thresholds. Ask Pi Agent to confirm the matching entities instead of assuming them.

  3. Request a plan before activation

    For any request that changes settings, keep the instruction to summarize the plan and wait for approval. Review the proposed change before it affects your home.

  4. Correct the result with a focused follow-up

    If the result is not right, state the difference: “It is too sensitive; require the condition to last for 5 minutes.” The current Session contains the earlier context, so you do not need to repeat everything.

Tip: If your request is incomplete, ask Pi Agent to identify missing details and clarify them with you before it proceeds.

Automations and scenes

Use these prompts to automate household routines. Each example defines a scope, exceptions, or an approval boundary.

Create a “Welcome home” scene. When the entrance sensor detects someone arriving, turn on the entry light, set the living-room air conditioner to 26°C, and play quiet music. Run it only after 18:00 on weekdays. First identify the entities you would use, summarize the configuration, and wait for my approval before enabling it. Respond in English.

Pattern: scene + schedule + approval before activation.

Create a “Good night” script that turns off all lights, sets the air conditioner to sleep mode, and checks that every door and window sensor reports closed. If anything is open, notify me and do not continue. Summarize the script and wait for my approval before enabling it. Respond in English.

Pattern: handle an exception with a notification instead of forcing the action.

When every household member’s phone has been away from home for more than 10 minutes, turn off the TV and the living-room power strip, then start the robot vacuum. If cleaning is still in progress when someone is approaching home, send the vacuum to its dock. Confirm the people and entities you would use, then wait for my approval before enabling the automation. Respond in English.

If the light in the child’s room is still on after 22:00, send a reminder. If it remains on at 22:30, lower it to 20% brightness. Exclude Fridays, Saturdays, and the school-vacation dates I provide. Summarize the logic and wait for my approval before enabling it. Respond in English.

Pattern: state exceptions up front to reduce rework.

Turn on the living-room dehumidifier only after humidity has remained above 70% for 30 minutes, and turn it off below 55%. Use appropriate hysteresis so it does not cycle near a threshold. Identify the sensors and controls, summarize the plan, and wait for my approval before enabling it. Respond in English.

Pattern: describe the desired behavior; let the agent propose appropriate hysteresis.

Use the washing-machine plug’s power history to detect when a cycle has finished: power should fall from its normal running range to near zero and remain there for 3 minutes. Notify me to unload the laundry, but delay overnight notifications until 07:00. Inspect the actual history before proposing thresholds, and wait for my approval before enabling it. Respond in English.

Turn on the balcony plant light 2 hours before sunrise and turn it off at sunrise. On consecutive overcast or rainy days, extend operation by 1 hour using my configured weather source. Explain how you will define those conditions and wait for my approval before enabling it. Respond in English.

Review my existing “Good night” automation. If the next morning’s calendar contains an event before 09:00, I want the wake-up scene to start 30 minutes earlier. Explain your proposed approach and identify any privacy implications of calendar access; do not change anything until I approve. Respond in English.

Pattern: explain the approach before changing an existing automation.

Dashboards and panels

Describe who will use the dashboard and what they need to see; Pi Agent can then propose the cards and layout.

Create a “Household energy” dashboard. Put current total power and today’s accumulated energy at the top, a daily bar chart for the previous 7 days in the middle, and today’s five highest-consuming devices at the bottom. Create it as a new dashboard view. Show me the proposed entities and layout before making changes. Respond in English.

Create a simplified dashboard for an older household member. Include only large controls for the living-room light, air conditioner, and TV; use large text and a tablet-friendly full-screen layout. Show me the proposed layout and wait for approval before creating it. Respond in English.

Create a “Before you leave” view for the entryway tablet. Show today’s weather, chance of rain, current outdoor temperature, and arrivals for the public-transit route and stop I specify. Confirm the official data sources and proposed entities before building it. Respond in English.

Reorganize the “Security” view. Put camera feeds at the top, list door and window sensor states below them, and clearly mark devices with batteries below 20%. Do not expose camera images to an external AI provider. Show me the proposed layout before changing the dashboard. Respond in English.

Create a “What happened at home today?” card that summarizes automation activity from the past 24 hours in plain language, one event per line, such as “18:32 — Welcome home scene started.” Show me which log or history data it will use before creating the card. Respond in English.

Data analysis and energy use

Ask Pi Agent to examine the relevant Home Assistant history, state its assumptions, and separate observations from estimates. Complex analysis may benefit from a reasoning model; see Chapter 12.

Analyze the living-room air conditioner’s energy use over the past 30 days. Compare periods set to 25°C with periods set to 26°C, account for weather and occupancy where data is available, and clearly separate observations from assumptions. Suggest a more efficient schedule, but do not change any settings. Respond in English.

Review the dehumidifier’s operation and indoor humidity over the past 2 weeks. Assess whether the current automation thresholds are reasonable and identify possible unnecessary operation. Explain the evidence and suggest revisions, but do not change anything. Respond in English.

My electricity provider uses time-of-use rates. Using the tariff schedule I provide, identify high-consumption devices that often run during peak periods. Suggest loads that could safely move off-peak and show every assumption behind any monthly cost estimate. Do not change schedules without my approval. Respond in English.

Draft a weekly household report for delivery every Monday at 08:00. Include energy use compared with the previous week and likely causes, devices below 20% battery, and automation failures with their current status. Confirm the notification service, recipients, data sources, and format before enabling it. Respond in English.

Using the past month of occupancy and light-state history, identify the three rooms where lights most often remain on while unoccupied. Explain data limitations and suggest where an occupancy sensor might help. Do not claim or invent savings; show only estimates based on the tariff and measurements I provide. Respond in English.

On the 15th of each month, estimate the current billing period’s electricity cost using the tariff and billing dates I provide. If the estimate is more than 10% above the previous comparable period, notify me and explain the main contributors. Show assumptions and wait for approval before enabling it. Respond in English.

Local information services

A suitable official or open-data feed can sometimes fill gaps in the Home Assistant ecosystem. Confirm the source, license, coverage, and update frequency before building an integration.

Use the official public-transit feed for my area to create a sensor showing arrivals for [route] at [stop], then place it on the entryway dashboard. Confirm the source, route and stop identifiers, update frequency, and field definitions before building it. Respond in English.

Use the official real-time feed for my local bike-share service to show available bicycles and empty docks at [station]. Change the dashboard indicator when fewer than 2 bicycles are available. Confirm the source, station identifier, and update frequency before building it. Respond in English.

I want a local waste-collection alert 10 minutes before the vehicle reaches my area. First find an authoritative public data source for my location, explain its coverage and limitations, and confirm the plan with me. Do not build or enable anything until I approve. Respond in English.

Pattern: verify an uncertain source before implementation.

Use my official local weather service. If tomorrow’s chance of rain exceeds 70%, notify me at 22:00 to bring in anything drying outside and take an umbrella when leaving. Confirm the forecast field, location, update time, and notification recipient before enabling it. Respond in English.

When the official emergency-management service issues a severe-weather warning for my area, or the relevant local authority announces a school or workplace closure, notify the household. Confirm the authoritative sources, jurisdiction, recipients, and exact trigger conditions before enabling it. Respond in English.

Create an exchange-rate alert for [base currency] and [quote currency]. Ask me for the threshold rather than assuming one, display the authoritative data source and timestamp, and let me change the threshold later. Show me the plan before enabling it. Respond in English.

Household knowledge base

Organize manuals, warranties, service contacts, and household notes so that you can retrieve them later. Do not add sensitive data unless its storage and access controls are appropriate.

Add this washing-machine manual PDF to the household knowledge base. When I ask about the washing machine, answer from this document, cite the page number, and say when the document does not contain the answer. Explain where the file and index will be stored before importing it. Respond in English.

Record the following details after confirming where they will be stored and who can access them: the water heater was installed in March 2024 by [installer name], has a 5-year warranty, and the service number is [service number]. Use this record for future warranty or repair questions. Respond in English.

Using the washing-machine manual in the household knowledge base, explain how to run the drum-cleaning program, what cleaning product the manufacturer permits, and how long the program takes. Cite the relevant pages and do not guess if the manual is unclear. Respond in English.

Create a warranty list for the appliances already recorded in the household knowledge base. Sort it by expiration date, with the earliest first, and flag warranties expiring in less than 2 months. Identify missing dates rather than inventing them. Respond in English.

Create a “Household maintenance record.” When I report a repair or consumable replacement, record the date, item, provider, and cost if supplied. Ask where the record should be stored and who may access it; never invent missing values. Respond in English.

Find the braised-pork recipe ratio I saved previously, cite the source note, and compare its ingredients with the household inventory. List anything missing as a shopping list, but clearly mark stale or uncertain inventory records. Respond in English.

Maintenance checks

Use these prompts to identify problems early, but require approval before repairs or configuration changes. For diagnosis guidance, see Chapter 22: Troubleshooting guide.

At 06:00 each morning, check which devices are unavailable, which batteries are below 20%, and whether any automations failed yesterday. Notify me only when something needs attention. Show me the entities, failure criteria, and notification recipient, then wait for approval before enabling it. Respond in English.

Pattern: notify only when attention is required.

The “Good night” automation did not turn off the living-room light yesterday. Review its trace and related logs, explain the likely cause, and propose a fix. Do not make any change until I approve; afterward, report exactly what changed and how it was tested. Respond in English.

I want to rename the “Entry sensor.” First list every automation, script, dashboard, and other reference that uses it. Show me a migration and rollback plan, and do not rename anything until I approve. Respond in English.

Check my automations for hard-coded device_id references. List each occurrence and explain whether changing it to an entity_id is appropriate; do not assume every case should be converted. Show me the proposed edits and wait for approval. Respond in English.

Check whether automatic backups have been running successfully and report the time, status, and available size information for the latest 3 backups. If there has been no successful backup for more than a week, ask for my approval before starting a full backup. Respond in English.

Working with Skills

Skills give Pi Agent task-specific procedures. See the Skill Handbook for selection and safety guidance, and Chapters 1416 for concepts, installation, and authoring.

Which Skills are loaded in this Session? Summarize what each one does and identify which, if any, contributed to your previous response. If the interface does not expose that information, say so rather than guessing. Respond in English.

Use skill-creator to draft a “Refrigerator inventory” Skill. When I provide a refrigerator photo, it should list visible ingredients, flag items whose supplied dates fall within 3 days, and suggest one dinner using available ingredients. Include privacy and food-safety caveats, then show me how to review and install it. Respond in English.

Turn my current weekly household-report procedure into a Skill. Preserve the existing data sources, format, and notification method, but first document them and identify any credentials or personal data involved. Draft the Skill for review; do not install it until I approve. Respond in English.

Evaluate whether this Skill is suitable for my home: git:github.com/tonylofgren/aurora-smart-home. Read its SKILL.md and related scripts first, summarize what it can do, list dependencies and permissions, and explain the risks. Do not install it until I approve. Respond in English.

Pattern: review a Skill before installing it.

Prompt-writing checklist

The examples follow four simple practices:

Say thisAvoid thisWhy
State the goal: “Remind me when the laundry is done.” Prescribe an unverified implementation: “Use a power threshold and a 3-minute delay.” A clear goal lets the agent inspect the available entities and propose an appropriate method.
Define scope and exceptions: “Run on weekdays except Friday.” Add exceptions only after building the automation. Stating exceptions up front reduces rework and unintended triggers.
Set an approval boundary: “Summarize the plan and wait for my approval before enabling it.” “Change it now.” A review step protects your live configuration.
Give focused feedback: “It is too sensitive; require the condition to last for 5 minutes.” Repeat the entire request. The current Session already contains the earlier context.

FAQ

Should I copy the prompts verbatim?
No. These examples demonstrate useful details: the goal, scope, exceptions, and approval boundary. Rewrite them in your own words and adapt them to your home.
What if I do not understand what Pi Agent created?
Say, “Explain in plain English what this automation does and exactly when it will run.” You can also request the same plain-language summary after every revision.
What if Pi Agent takes the wrong approach?
Explain what is wrong and ask it to revise only that part. You do not need to repeat the entire request. Requiring a summary and approval before activation catches many mistakes early. If you are still stuck, see the Chapter 22 Troubleshooting Guide.
How much will these prompts cost?
Cost depends on the provider, model, context size, and tool calls. A broad analysis of 30 days of history will generally use more tokens than a short status request. Check current provider pricing and see Chapter 12 for model-selection guidance.
Can I share this library with my household?
Yes. Download this page as a single HTML file and share it through your preferred channel. Each person can copy and adapt the relevant prompts.

Download this handbook

This handbook is a self-contained HTML file with embedded icons. Download it for offline use or share it with your household.

Download handbook HTML Return to the Resource Hub