local-service-research - SKILL.md Agent Skill

name: local-service-research description: > Research and compare local service providers — any category: massage/spas, contractors, plumbers, electricians, vets, daycares, salons, cleaners, dentists, mechanics, movers, etc. Gathers reviews, ratings, and pricing from the web, writes a structured note per place, and rolls everything into a single shareable comparison report. Use this whenever the user is evaluating, vetting, or choosing between local businesses — e.g. "is X any good?", "find me a good Y near me", "compare these places", "should I go here?", "check reviews for...", or when they paste a list of local business names to assess. Trigger even if they don't say the words "research" or "compare" — any time they are deciding among local providers, this skill applies.

Local Service Research

Help someone decide which local business to use. The job is to turn a name (or a list of names, or "find me a good X near Y") into a clear, sourced recommendation — and to keep that work organized so it can be revisited and shared.

People making these decisions are usually busy and not looking for a wall of text. They want: is this place good, what does it cost, and should I go? Lead with the verdict, back it with evidence, and flag anything that should give them pause.

When to use this

The user names a specific local business and asks if it's any good.
The user pastes a list of candidate businesses to vet.
The user asks you to find good providers of some service in an area.
The user is comparing options and wants pricing or reviews.

It works for any local service category. Massage, plumbing, childcare, auto repair — the workflow is identical; only the criteria shift.

Workflow

1. Establish the project folder (first time only)

These research notes are worth keeping, so save them to files rather than only printing to chat. On the first relevant request in a session, ask the user where to save — or offer to create a project folder in the current working directory (e.g. ./<topic> Research/). Don't assume a path; different people have different setups. Once established, reuse it for the rest of the session.

If the user already pointed you at a folder earlier, just use it.

2. Gather candidates

If the user gave names, research those.
If the user asked you to find providers, search the web (Google Maps, Yelp, the category's directories) for well-reviewed options in their area first, then present a shortlist before deep-diving.

3. Research each place

For every business, search the web and pull together:

Identity & logistics — address/area, phone, hours, booking method, parking, website.
Ratings — star rating and review count, ideally from more than one source (Yelp, Google, Birdeye, category-specific sites). Note the count, not just the score: 4.9 from 8 reviews is weaker evidence than 4.4 from 300.
What people praise — recurring positives, named staff if reviewers consistently single someone out.
Cons / red flags — recurring complaints. Distinguish a one-off gripe from a pattern. Genuinely concerning signals (safety, billing, no-shows, licensing) get surfaced plainly, not buried.
Pricing — concrete prices for the service the user cares about. Note the fine print (cash-only, card fees, new-customer discounts, membership). If you can't find real prices, say so and suggest calling — don't invent numbers.
Tier / category fit — sometimes options aren't the same kind of thing (e.g. a budget walk-in shop vs. a licensed clinic). Call that out so the user compares like with like.

Save one note per place using the template in assets/research-note-template.md. See references/sources-and-scoring.md for which sources to trust and how to weigh ratings.

4. Write the verdict

End each note with a plain-language verdict: who it's best for, and whether to go. Hedge only where the evidence is genuinely thin.

5. Build the comparison — readable first, shareable second

Once there are two or more places, the user needs to review the answer, not open a file to find it. So work in this order:

Present the comparison inline in the conversation. Lead with the verdict (who to pick and why), then a Markdown comparison table, then the "quick takeaways" (cheapest, best-rated, best for the user's specific need, anything to confirm by phone). This is what the user actually reads — don't bury it in an attachment.
Save a Markdown comparison file (<topic> Comparison.md) alongside the per-place notes. Markdown is the right default because the user can read it in any editor, in the chat, or on their phone — no browser required.
Only then, optionally, generate the HTML report from assets/report-template.html — and only when the user wants something shareable/printable (to email, or to Print → Save as PDF). Treat HTML as an export format, not the place the user reviews the work. If you're not sure they need it, offer it rather than producing it unasked.

This skill does not auto-generate PDFs, to avoid depending on tools the user may not have installed — the HTML export covers that need on demand.

Research note structure

Use this shape for each per-place file (full template in assets/research-note-template.md):

# <Business> (<area>)
**Researched:** <date> · **Category:** <service> · **Status:** <verdict in 3 words>

## Details         — address, phone, hours, booking, parking, website
## Ratings         — score + review count, per source
## Pricing         — concrete prices for the relevant service + fine print
## Pros            — recurring positives
## Cons / red flags — recurring complaints, concerns surfaced honestly
## Verdict         — who it's for, go / skip

Name files so they sort and scan well, e.g. YYYY-MM-DD <Category> - <Business>.md.

Reporting style

Verdict first. People want the answer, then the reasoning.
Cite sources. Link the review pages you used so the user can verify.
Be honest about evidence strength. Few reviews, no pricing, or a single loud complaint — say so rather than projecting false confidence.
Respect the user's stated needs. If they care about neck/shoulder massage, or weekend availability, or pet-friendliness, weight the comparison on that.

Honesty and limits

Don't fabricate ratings, prices, or quotes. If a source is blocked or thin, report what you actually found and where to confirm.
Review sites can be scraped or gamed; a brand-new 5.0 with a handful of reviews deserves a note of caution, not a crown.
Surface real red flags (licensing, safety, billing disputes) clearly — this is the part of the research that protects the user.

Reference files

assets/research-note-template.md — per-place note template.
assets/report-template.html — self-contained comparison report (print to PDF).
references/sources-and-scoring.md — where to look, how to weigh ratings, category-specific criteria.