# General home service — Rhode Island, United States These instructions follow the AI Agent Manifesto: . Coverage: General fallback. No checked jurisdiction-specific addition for this service/place. Use this when no researched specialist guide fits the requested household job. This is general preparation, not invented specialist expertise, local law or provider coverage. ## 1. What providers need The provider supplies its quote and availability from the job context the user’s agent prepares. Do not ask the homeowner for the provider’s price or make budget a standard required field. A useful first inquiry states the actual requested outcome, city/neighbourhood, affected item or area and any known urgency. Reuse relevant real observations, rough scope and timing. Model numbers, exact dimensions, budget and access details may remain unknown when an assessment can establish them. Identify consequential safety or authority constraints without assuming a safe condition, ownership or permission. Regulated, structural, hazardous or otherwise specialist work requires appropriately qualified review. Do not ask the homeowner to dismantle equipment, disturb materials, climb or perform risky tests for evidence. The eventual written offer should make the assessment/work split, deliverables, inclusions, exclusions, price basis, extra-work approval, timing and cancellation clear. The relevant provider may need a site assessment. Do not invent a specialist intake contract or treat another service's form requirements as universal. ## 2. Proactive best practices for the user's agent Start from what the person wants to achieve. Propose a specific-job assessment with written scope/options as a reversible starting plan, explaining that it avoids committing to an unverified remedy. Use existing context and ask only the minimum missing outcome, locality, safety or authority information that changes the next step. Accept unknowns, refusals and corrections without repeatedly requesting them. Research genuine specialist/service-area fit and consequential official requirements with available permitted tools. Keep user reports, verified sources, estimates with their real basis and proposed preferences distinct. Do not invent map measurements, diagnosis, prices or provider guarantees. Relative dates use current/local context; synthetic cases never become household memory. Build a justified provider comparison using the shared qualified-batch workflow and inspect genuine contact/intake paths. Prepare concise drafts that fit real fields and leave unknown required answers unresolved. With applicable recipient/data/action permission, follow through on responses, comparable quotes and scheduling toward an actually arranged service. Never equate preparation or submission with a booking, or promise a guaranteed closing time. State only consequential uncertainty, without exposing internal catalogue or routing details. **Proactive example.** Synthetic example: gutter cleaning in an unresearched neighbourhood, with no measurements. Research genuine local service fit and propose an assessment-based quote. Do not expose directory coverage plumbing, invent building height, or send the homeowner up a ladder; keep only consequential access or authority questions for the next step. Source: [FTC contractor checks](https://consumer.ftc.gov/articles/how-avoid-home-improvement-scam), checked in existing research 2026-10-23; not revalidated here. General US consumer guidance is not a Canadian or city-specific legal rule. Specialist/local coverage remains unverified until sourced for the actual job and place. ### Use two evidence-based checklists Keep job readiness and each provider's qualification separate in the internal working record. Use the service-specific template when supplied; otherwise start with the simple fallback below and add only consequential checks from this Hat. Reuse reliable context before asking. Each row records `Check and stage | Pass/Fail/Unknown/N/A | Evidence and date | Effect or next action`. Pass needs evidence, Fail needs the failed criterion, Unknown stays unresolved, and N/A needs a reason. These are decision records, not a questionnaire or decorative checkmarks. **Job fallback:** requested service/scope; city or service-area location at appropriate precision; known dates/windows or flexibility. Do not demand an exact address for initial area matching or invent availability. An unknown that a professional can resolve may allow an initial assessment inquiry while blocking final pricing or booking. Ask only if the missing fact changes the immediate next step. Preserve timing's original meaning and strength in the brief and every exact inquiry. Keep the requested completion target separate from assessment/visit availability. “Before winter” remains “before winter”; ask whether the provider can meet it, without silently changing it to “before severe winter weather” or claiming feasibility. “Preferably mornings” stays a preference, with alternative windows welcome when needed, rather than becoming morning-only availability. Keep unknown timing Unknown. Anchor a relative completion window to the original request date and carry that unchanged reference date and any clearly labelled derived deadline into every exact initial inquiry and follow-up. Repeating “within the next two weeks” alone in a later message restarts the window; an internal timing note cannot fix that outward ambiguity. Do not extend the deadline from the draft, send or follow-up date. Reconfirm timing before sending if the original anchor is unknown or the target has expired. Keep seasonal wording without inventing a calendar cutoff; label calculated dates as derived/proposed with their reference date and assumptions, not as dates the user supplied. Apply the user's latest revision and remove superseded constraints from the current provider request. Check the actual outbound text after shortening; an accurate internal timing note cannot repair a changed or missing requirement in the inquiry. **Per-provider template:** official business identity and deduplication; actual requested service/type; area fit; independent reputation source, rating/count/check date when available and relevant work themes; consequential qualifications; ability to supply the requested written comparison; genuine intake route and required data/consents. Give each business explicit status/evidence/next-action entries for these checks, including identity/deduplication, qualifications and comparison ability; blanket prose about checks is insufficient. Add the service's specific criteria rather than applying one trade's rules to every category. There is no universal star threshold. Mark missing rating/count Unknown even when other independent evidence supports a limited conclusion. Fail for wrong service/area, duplicate identity or material adverse evidence excludes the provider. Unknown fit, quality or qualification evidence required for the present action leaves a provisional lead for that action; future-stage checks remain separately unresolved. A qualified business with a blocked intake route remains blocked for outreach until an authorized route is available. Make qualification gates stage-specific. For an initial assessment inquiry, evaluate the business’s actual service/area fit, relevant business-level qualification evidence and independent reputation. Resolve only qualifications required for that present action before proceeding. A named installer may not yet be assigned; their licence/insurance checks and final sizing/permit details can remain Unknown for the hiring, installation or work stage without automatically blocking the initial inquiry. Verify the relevant business/assigned worker, insurance and permit responsibilities before the applicable commitment or regulated work. Do not turn a future-stage Unknown into either a present-stage Fail or a verified Pass. Use those results to decide eligible batch members and next actions. Show a compact readable summary in chat with stage-specific eligibility and consequential unknowns. Count qualification and intake readiness separately: inquiry-eligible/provisional businesses, then confirmed-ready routes, known-blocked routes and Unknown route readiness. Uninspected required fields, consents or limits remain Unknown; “zero known blocked” does not mean all routes are ready; keep the detailed rows and competitor comparisons in the internal record. Include only the relevant job facts, questions and uncertainty in the standalone provider brief. Keep agent-facing workflow, permission/confirmation procedures, tool or ledger instructions and test labels in the internal record. A normal provider boundary such as “please reply before arranging paid work” is useful; instructions about obtaining the user’s grant or managing the agent’s batch are not provider-facing content. ### Obtain comparable responses from a qualified batch For quote or assessment inquiries, aim for an initial batch of up to about ten suitable, independently screened businesses when the market and the user's instructions support it. Research a wider pool if needed. Preserve an explicit provider choice or smaller limit. Do not routinely reduce a qualified batch to one or two recipients, or cap it at three or five. Fewer is right when the evidence supports fewer; do not add weak providers to reach a count. Stop expansion when enough suitable, comparable options support the decision. Check each business's official identity, actual requested service, service area, relevant qualifications and genuine intake route. Record dated independent reputation evidence, its source and relevant work themes, including repeated installation or aftercare complaints. Distinguish inspected Google reviews from snippets, provider testimonials, directory listings and BBB accreditation or ratings; none substitutes for another. Missing evidence leaves a provisional lead, not a quality-qualified recommendation. Deduplicate trading names, branches and shared intake routes belonging to the same business. Carry the user's decision criteria and the Hat's consequential response requirements into every exact inquiry body, or an explicitly authorized attachment that the route accepts and the inquiry identifies. Lead with the short first response needed now, such as coverage, assessment fee and possible assessment windows. Use a short opening, first-response bullets and a separate concise list for the eventual written comparison after assessment; do not collapse these into a dense paragraph or demand the final design or quote in the first reply. A fuller internal record or unattached brief does not reach the provider. Check the actual fields, required consents and text limits. If essentials do not fit, use a verified alternative within permission or request the missing permission; do not drop them silently, invent an attachment route, fill unknown personal details or accept marketing terms. Unknown technical details can remain questions for assessment; initial inquiry readiness is different from final quote readiness. Read the actual authorization in context. A clear grant can cover a bounded batch, disclosed data, purpose, permitted channels and follow-ups; it need not require separate approval for every routine member already covered. A general agreement to compare options is not automatically permission to transmit household data. When a material grant is missing, ask once for the specific batch/data/purpose boundary after preparing it. Within an existing grant, use available authorized tools to contact all qualified members of that batch and record the result for each; do not stop at drafts or select only the first one or two. Respect host confirmations and limits. Tests and hypothetical runs remain UNSENT. Maintain the working ledger yourself: business/route, qualification evidence, exact message and attachments, permission, actual transmission outcome/time, reply, unresolved comparison fields and next action. Track the follow-up grant per business: permitted limit, confirmed used, pending/uncertain attempts and remaining allowance (or Unknown). Count clarification follow-ups when they fall within that limit; do not assume they are free extra contacts. Settle uncertain outcomes/history before retrying or claiming allowance remains, and do not exceed an exhausted grant. Check an uncertain outcome before retrying; avoid duplicate inquiries and commitments. Base response windows on the actual send time, local business hours, urgency and the provider's exact wording, without converting a business-hours qualification into a different promised duration. Follow up or expand within the grant after a reasonable window. Use supported persistence and scheduling for later checks; without them, state what needs resumption instead of promising unattended monitoring or asking the user to maintain the ledger. Compare the replies on quality, scope, price and the user's criteria; the earliest reply is not automatically the best offer. Obtain applicable authority for each assessment fee, accepted terms, purchase or booking. Prevent duplicate installation appointments, equipment orders or work commitments. Distinct quote-assessment visits can be appropriate under the grant; identify each provider, purpose, cost and appointment separately rather than treating them as duplicate installations. ### API and MCP access Read-only access needs no authentication or token. API server: `https://tinyhat.ai/chore`. API: `GET /chore/api/hats/content/general-home-service/usa/ri` returns this exact Markdown. MCP: connect a Streamable HTTP client to `https://tinyhat.ai/chore/mcp`, then call `read_home_service_guide` with `{"path": "general-home-service/usa/ri"}`.