name: client-research description: "Use during Stage 2 of the website factory pipeline to autonomously harvest Google Business Profile, all reviews, owner profile, services/offerings, service areas, competitor landscape, brand identity signals, business model, niche-specific certifications, special offers, financing/pricing, and trust signals into research-report.md plus research.json. Read in full before invoking the research-agent. Output is the spine that every downstream stage reads. Uses tools/apify-scrape.py for reliable scraping (Google Places + Facebook + website crawl + Instagram + niche-specific directories)." allowed-tools:

• Read
• Write
• Edit
• Bash
• WebFetch
• WebSearch

metadata: version: "1.0.0"

Client Research Skill

How this skill is customised

This skill defines the universal research-data.json shape and the universal scrape mechanism. Every niche needs the same core fields: business name, owner, address, service areas, reviews, social URLs, competitor landscape.

The niche-specific extensions, which certifications/affiliations matter, which discount audiences are common, which financing/pricing patterns exist, which directories to search (BBB for contractors; TripAdvisor / The Knot for hospitality; Houzz for designers; etc.), which signal patterns indicate business model and market segment , come from the niche playbook:

templates/{active-niche-slug}/niche-playbook/research-extensions.md
templates/{active-niche-slug}/niche-playbook/research-extensions.schema.json

research-extensions.md is a markdown brief Module 2D writes describing what additional fields and signal patterns to look for. research-extensions.schema.json extends the universal research-data.schema.json with niche-specific fields that get nested under research-data.json.nicheExtensions.{slug}.

If the niche extension files are missing, the agent halts with a Module-2D pointer.

Purpose

Produce the complete research artifact that every downstream stage depends on. The audit reads it. The asset scraper reads it. The strategy reads it. The copy deck reads it. The template matcher reads it. The hero image reads it. The proposal reads it. This is the spine of the pipeline. Be thorough.

Outputs (two files)

clients/[Client Name]/Pipeline Data/research/research-report.md

Long-form narrative report. The universal sections (always present):

1. Business Overview
2. Services / Offerings
3. Service Areas / Geography
4. Owner / Founder / Host Profile
5. Vision, Mission & Values
6. Brand Identity Signals
7. Brand Voice & Copy Tone Analysis
8. Business Model & Market Segment
9. Reviews & Reputation
10. Trust & Certification Signals
11. Competitor Landscape
12. Tech Stack & Current Site Assessment
13. Differentiators & Unique Story Hooks

The niche playbook may add or rename sections (e.g. hospitality adds "Occupancy & Seasonality"; contractor renames "Services" to "Services Offered"). The agent applies the niche playbook's section overlay after writing the universal sections.

clients/[Client Name]/Pipeline Data/research/research-data.json

Structured payload for downstream agents. Universal shape:

{
  "businessName": "",
  "websiteUrl": "",
  "phone": "",
  "phoneNormalised": "",
  "email": "",
  "yearFounded": "",
  "yearsInBusiness": 0,
  "ownerName": "",
  "ownerBio": "",
  "ownerLinkedIn": "",
  "ownerValues": [],
  "visionStatement": "",
  "missionStatement": "",
  "primaryCity": "",
  "state": "",
  "fullAddress": "",
  "serviceAreas": [],
  "primaryService": "",
  "services": [],
  "businessModel": {
    "primaryFocus": "",
    "secondaryFocus": ""
  },
  "marketSegment": {
    "primaryFocus": "",
    "secondaryFocus": ""
  },
  "googleReviewCount": 0,
  "googleRating": 0.0,
  "googleReviewUrl": "",
  "facebookReviewCount": 0,
  "facebookRating": 0.0,
  "facebookUrl": "",
  "thirdPartyReviews": {},
  "certifications": [],
  "specialOffers": {
    "items": [],
    "discountAmount": ""
  },
  "pricing": {
    "model": "",
    "lowEnd": null,
    "highEnd": null,
    "currency": "",
    "termsDescription": ""
  },
  "brandGuidelines": {
    "hasFormalGuide": false,
    "guideFilePath": "",
    "lockedColors": {},
    "lockedFonts": {},
    "lockedTagline": "",
    "lockedVoiceRules": []
  },
  "competitors": [],
  "brandVoice": "",
  "brandVoiceTags": [],
  "primaryColorGuess": "",
  "secondaryColorGuess": "",
  "tagline": "",
  "logoStyle": "",
  "hasMascot": false,
  "mascotDescription": "",
  "techDifferentiators": [],
  "uniqueStoryHooks": [],
  "techStack": "",
  "currentSiteIssues": [],
  "nicheExtensions": {}
}

nicheExtensions.{slug} contains the niche-specific fields the niche playbook adds. For example, a contractor playbook adds nicheExtensions.contractor.manufacturerCerts, nicheExtensions.contractor.businessModel.retail/insurance, and nicheExtensions.contractor.specialOffers.veterans/seniors/firstResponders. A hospitality playbook adds nicheExtensions.hospitality.occupancyPattern, nicheExtensions.hospitality.weddingEnabled, nicheExtensions.hospitality.directBookingFlag.

Step-by-step execution

Step 1, Read niche extension

Load templates/{active-niche}/niche-playbook/research-extensions.md to learn what additional signals to detect, what directories to search, what brand-archetype tags apply. Halt if missing.

Step 2, Fetch the client's website

Playwright (mobile + desktop) capture: full HTML of homepage, all H1/H2/H3, all phone numbers, all emails, address, services/offerings, social links, review widgets, tech stack signals, page load timing, mobile responsiveness, blog presence + last post date, image alt text, schema markup. Save raw HTML and screenshots.

Step 3, Crawl internal pages (depth 1)

About, services/offerings, contact, reviews, blog index, careers/team. Extract title, H1, word count, key phrases per page. Save to internal-pages.json.

Step 4, Google Business Profile lookup

Web search: "{businessName}" "{primaryCity}" Google Business Profile site:google.com. Extract verified address, service area, categories, years in business, review count, star rating, photo count, hours, direct review-tab URL.

Step 5, Facebook lookup

Web search: "{businessName}" Facebook reviews. Extract URL, review count, rating, post cadence.

Step 6, Third-party directory lookups

For each directory the niche playbook lists in research-extensions.md (BBB, TripAdvisor, The Knot, Houzz, Angi, Yelp, etc.), web-search and extract: profile URL, rating, review count, accreditation/membership status, years listed. Populate thirdPartyReviews with one entry per directory.

Step 7, Owner / founder / host identification

Web searches in this order:

1. "{businessName}" owner founder
2. "{businessName}" "about us"
3. "{businessName}" CEO president host
4. "{businessName}" LinkedIn

Extract: full name, bio/background, years in the niche, family/military/heritage/local angles, LinkedIn URL, photo URL.

Look explicitly for verbatim vision/mission/values text and capture exactly as written.

Step 8, Service area discovery

Combine website "service areas"/"locations" pages + GBP service area + footer cities + schema areaServed. Output a clean array with primary city flagged, state, approximate population.

Step 9, Competitor identification

Web search for top local competitors using the niche playbook's keyword templates. Extract top 5 per niche-relevant query. For each: business name, website, review count, star rating, years in business, notable differentiator.

Step 10, Brand identity + voice signals

From website + visible asset photos + social media:

• Primary brand colour (best guess from homepage hero/logo dominant)
• Secondary brand colour
• Tagline if present
• Logo style: wordmark | mascot | monogram | emblem | icon-with-text | script
• Has mascot? Mascot description if present
• Brand voice across: tone, formality, confidence, local emphasis, heritage emphasis
• 2-3 sentence brand voice summary; 3-5 brand voice keywords

Step 11, Niche-specific signal extraction

Apply the patterns from research-extensions.md to detect niche-specific signals. Populate nicheExtensions.{slug} with the structured payload defined in research-extensions.schema.json.

For example:

• Contractor niche: manufacturer certifications (GAF/CertainTeed/etc.), business model

(retail/insurance/both), market segment (residential/commercial/both), special-audience discounts (seniors/veterans/first-responders), financing (0% available, providers).

• Hospitality niche: occupancy pattern (year-round/seasonal), wedding-enabled flag,

direct-booking flag, accommodation types, capacity, dining/spa amenities.

• E-commerce niche: product categories, free-shipping threshold, return policy, brand-

archetype signals.

Never fabricate. Each detected field must point at a real source (URL or fetched-from-page).

Step 12, Brand guidelines check

Check clients/[Client Name]/Pipeline Data/brand-guidelines/ for a user-supplied brand kit. The intake optionalOverrides.brandGuidelinesPath may also point to a file. If present, extract locked colours, fonts, tagline, voice rules. A formal brand guide always wins over inference.

Step 13, Tech stack lightweight assessment

CMS detected, HTTPS, schema markup presence, sitemap.xml/robots.txt presence, mobile viewport tag. Stage 3 does the deep audit.

Step 14, Compile + log

Assemble research-report.md with all universal + niche-overlay sections. Assemble research-data.json with universal fields + populated nicheExtensions.{slug}.

Append to logs/build-log.md:

[ISO_TIMESTAMP] Stage 2 (Research), PASSED, {businessName}
  - Years in business: X
  - Service areas: N cities
  - Google reviews: X (rating Y)
  - Niche signals: {niche-specific summary from research-extensions}
  - Has mascot: yes/no
  - Brand guide supplied: yes/no
  - Brand voice: [summary]

Update pipeline-state.json → stages.research.passed = true.

Failure modes

Quality bar

• Every section has at least 100 words of substance
• Every claim has a real source (URL or fetched-from-page)
• Numbers are exact, not approximated
• Owner story is at least one full paragraph if found
• Brand voice analysis is opinionated, not generic
• Differentiator hooks are specific, not "they offer quality service"

If any section is sparse, either dig harder or explicitly mark "Insufficient data", don't pad with filler.

What this skill is NOT

• Not the SEO audit (Stage 3)
• Not the asset scraper (Stage 4)
• Not the strategy generator (Stage 5)
• Not the copywriter (Stage 6)
• Not the deep social-resonance miner (Stage 15)

This stage is purely about gathering and structuring information. Other stages do things with it.