---
name: author-home-service-guides
description: Write and evaluate concise, proactive service/location instructions for the user's agent. Use when authoring or improving a Tinyhat household-service guide. Do not use as a homeowner questionnaire or permission to contact providers.
---

# Make the next useful step easier

These instructions follow the [AI Agent Manifesto](https://tinyhat.ai/chore/manifesto).

Write for the **user's agent**, whose job is to move the chore toward a suitable quote and an actually arranged service with less homeowner effort. The user supplies relevant job/context information; the **provider supplies its quote and availability**. Never ask the user to supply the provider's price. Budget is not a mandatory standard intake field.

## A small improvement loop

**Author.** Decide what would let a suitable provider give the next useful response. Write two concise Markdown sections: what providers need, and proactive best practices for the user's agent. The first is an internal checklist, not a script of questions. Distinguish details needed for the next inquiry from details that can wait for assessment or booking.

**Try.** Walk through realistic requests with the guide, using available authorized test surfaces. Pick cases that challenge its judgement: perhaps known context, an unknown measurement, a refusal, an unresearched place, or a consequential service boundary. Record what the agent would do and the few things it would still need from the user. A written walkthrough is not a live model/host test.

**Critique and revise.** Ask whether the agent could have done more useful work before interrupting the homeowner. Remove questions that repeat context, can be researched, or do not change the next step. Repair unsupported assumptions and test the affected case again. Retain useful service-specific knowledge rather than adding rules for every example.

## What good judgement looks like

Start from real context and current timing. Research genuine service-area fit, relevant requirements and actual intake paths with available permitted tools. Offer a reversible starting plan with a brief reason and room for correction. Defaults are proposed preferences, not facts about equipment, safety, ownership, provider terms or price. Unknowns and refusals should usually change the plan, not restart a questionnaire. Essential safety/authority questions remain when they change the immediate route.

Cite local sources with jurisdiction, applicable work and date. Use general best practice when no checked local addition exists; inherited regional evidence is not city validation. Keep user reports, verified facts, estimates and proposals distinct. Do not invent research, map measurements, reviews, credentials or availability.

Help the user's agent progress through real provider comparisons, concise form-specific drafts, quotes and scheduling under applicable permissions. Inspect actual required fields and limits; leave unresolved answers unresolved. Known personal information is not permission to transmit it, and entering a live form can transmit before Submit. Preparation, an inquiry and a provider-confirmed appointment are different outcomes.

Apply the shared [provider workflow](provider-workflow.md), included in every generated Hat. Test an initial qualified batch of up to about ten, evidence-based fewer, and contacting the whole batch under an existing bounded grant. Check that every exact inquiry retains the user's decision criteria or identifies an authorized attachment the route accepts. Evaluate reputation provenance, technical unknowns, dated evidence, blocked routes, grant reuse, nonresponse and duplicate commitments. Critique the actual saved messages and simulated progression, repair failures and rerun affected cases until the intended standard is met or a concrete blocker remains. Keep expected and observed outcomes per case; do not call a written walkthrough or mock-send result live outreach.

Supply two usable checklist layers: service/job readiness and qualification of each provider. Tailor consequential checks to the actual service/type, with Pass/Fail/Unknown/N/A, evidence/date and the next decision. Record each business's identity/deduplication, qualifications and comparison ability explicitly rather than only in blanket prose. Where no specific template exists, start with requested service/scope, area at the needed precision and known timing/flexibility. Do not turn an unknown model, exact address or schedule into a universal first-inquiry blocker. Test that failed service/area fit and unknown quality evidence affect batch eligibility, while professional-confirmable technical unknowns remain honest. Use concrete service/scope fixtures, scannable staged inquiries and the distinction between authorized separate assessments and duplicate installation commitments. Require a compact status summary, not a long homeowner questionnaire or unexplained checkmarks.

Keep a short before/after record of the actual instruction change and its evidence. The standard is useful progress with less user work—not a larger checklist, a complete data packet or a guaranteed closing time. Keep this authoring skill separate from the guides.