local-lead-hunting

name	Local Lead Hunting
description	Run a repeatable, browser-first local prospecting workflow that finds real businesses, checks their websites, saves evidence, and separates strong leads from weak ones.
emoji	🧩
version	2.2.0
triggers	find local leads, local lead hunting, google maps leads, go through google maps for leads, scan businesses in, scan local businesses, prospect frederick businesses, prospect local companies, audit local business websites, check businesses for website problems, qualify local companies, build a local lead list, save evidence while researching, local seo opportunity, website lead qualification

Local Lead Hunting

Run a repeatable, browser-first local prospecting workflow that finds real businesses, checks their websites, saves evidence, and separates strong leads from weak ones.

---

When to Use This

Use this skill when the user wants real local-business prospecting, not generic web-research summaries. The work should begin from live business listings — preferably Google Maps — then move into the business website, then into qualification.

This skill is for questions like:

• Which local businesses in this city look like good leads?
• Can you go through Google Maps and check actual business websites?
• Which businesses have weak websites, weak SEO, or poor conversion setup?
• Can you save all the evidence while you research so nothing gets lost?

Do not use this skill for generic market research, competitor writeups, or one-off web searches with no lead qualification goal.

---

Core Principle

Map first. Website second. Qualification third. Deep analysis fourth.

Do not start by scraping random search snippets and guessing. Start from real local listings, confirm the business is real, inspect the site, record evidence, and only then decide whether the opportunity is worth deeper analysis or outreach.

Use browser tools for the listings/map workflow and live site inspection. Use web_search({ fetch_top_k }) or web_fetch_batch for supporting research such as reviews, directory pages, competitor pages, and multiple candidate websites that do not need interaction.

The workflow must be persistent. Save findings to workspace files as you go. Do not leave important findings only in chat.

---

Scope Boundary

This skill SHOULD trigger for

• local lead generation for Xpose Market or similar agency work
• city-based prospecting from Google Maps or local listings
• scanning local businesses and checking their sites
• qualifying businesses based on website / SEO / conversion weaknesses
• saving lead evidence and audit artifacts during research

This skill should NOT trigger for

• generic SEO advice not tied to lead-finding
• broad market research with no business-by-business qualification
• cold outreach writing by itself
• enterprise technical SEO audits with no prospecting component
• casual questions about a city or business category

---

Required Workspace Persistence

For every lead-hunting run, create or update a stable research folder. Example:

lead-research/xpose-market/
  README.md
  prospects.csv
  YYYY-MM-DD-<city>-run.md
  evidence/
  reports/

Minimum persistence requirements:

1. Run log

   • Create a dated markdown file for the session or update the current one.
   • Record geography, niches scanned, businesses checked, top candidates, and disqualification reasons.

2. Master prospect file

   • Maintain a cumulative CSV or table with one row per business.
   • Update existing rows instead of duplicating businesses on revisits.

3. Per-business evidence

   • For promising leads, save a dedicated evidence file with URLs, observed issues, qualification notes, and timestamps.

4. Audit output linkage

   • If a deeper analysis report is generated, write the report path and a short summary back into the prospect record and run log.

If the user says no data should be lost, treat these file writes as mandatory.

---

Standard Data Schema

Capture these fields whenever available:

Field	Meaning
business_name	Business name
category	Restaurant, salon, contractor, dental office, etc.
city	Target city / area
listing_source	Google Maps, local pack, directory, etc.
maps_url	Listing URL
website_url	Business website
has_website	yes / no
website_quality_score	1-10 quick score
seo_basics_score	1-10 quick score
conversion_score	1-10 quick score
local_seo_opportunity	1-10 score
viability_score	1-10 score
status	high / medium / low / disqualified
primary_issues	Main problems observed
outreach_angle	Best outreach hook
recommended_service	website rebuild / local SEO / CRO / lead gen
evidence_file	Path to evidence note
audit_report	Path to saved deeper report
last_checked	Date checked

Suggested CSV header:

business_name,category,city,listing_source,maps_url,website_url,has_website,website_quality_score,seo_basics_score,conversion_score,local_seo_opportunity,viability_score,status,primary_issues,outreach_angle,recommended_service,evidence_file,audit_report,last_checked

---

Qualification Framework

High potential

Use when most of these are true:

• real and active local business
• meaningful website, SEO, or conversion weakness is visible
• category where calls, bookings, leads, or foot traffic matter
• believable value proposition for Xpose Market
• outreach angle is specific, not forced

Medium potential

Use when:

• business is legitimate and relevant
• some weaknesses are visible
• opportunity exists, but urgency, budget fit, or leverage is less obvious

Low potential

Use when:

• business is real but improvement opportunity is modest
• website is acceptable already
• there is no strong hook for outreach

Disqualified

Use when:

• duplicate, closed, irrelevant, or bad-fit business
• franchise / edge case where outreach is weak
• site and positioning are already strong enough that the pitch would feel forced

---

Operating Workflow

Phase 1 — Prospect discovery

1. Open browser tools and go to Google Maps or another live local business listing source.
2. Search by geography plus niche when helpful.
3. Visually scan actual business listings.
4. Capture raw candidates quickly.
5. Skip obvious non-fits early.
6. Before manual scrolling loops on listing pages, make one real page interaction first when practical (for example clicking into the listings/results pane or selecting a business card). This avoids automation-layer blind-scroll guards that can block repeated browser_scroll(...) calls on some sites.
7. If the goal is broad candidate harvesting rather than tight UI steering, prefer bulk collection/extraction patterns over repeated single-scroll retries.

Phase 2 — Quick qualification

For each candidate:

1. record the listing URL
2. note category, reviews, legitimacy, and whether a website exists
3. open the website if available
4. score first-pass quality using:
   • mobile feel
   • design trust
   • CTA clarity
   • contact accessibility
   • messaging clarity
   • obvious technical / SEO gaps

For batches of candidate website URLs, use web_fetch_batch to collect first-pass text evidence quickly before deciding which sites deserve browser inspection or deeper analysis.

Phase 3 — Deep analysis for promising leads

Only for stronger candidates:

1. run a deeper website analysis
2. use deploy_analysis_team when the site is worth a fuller breakdown
3. save the resulting report path
4. extract the most useful findings for future outreach

Phase 4 — Final qualification

Assign one of:

• high potential
• medium
• low
• disqualified

Then update the master prospect file and run log.

---

Using deploy_analysis_team Well

deploy_analysis_team should be used on selected promising sites, not every business in the list.

Use it when:

• the business is real and relevant
• the site exists
• the business appears to have genuine upside
• a deeper report would help future outreach or prioritization

Recommended operating pattern:

1. Main Prom agent scans local listings and websites
2. Main Prom agent saves quick-screen findings immediately to workspace files
3. For strong candidates, run deploy_analysis_team(url)
4. Save the returned report path and key findings into the prospect record
5. Continue scanning other businesses while stronger candidates accumulate reports

This keeps discovery moving while deeper analysis strengthens the best leads.

---

Xpose Market Workflow

For Xpose Market revenue prospecting (or similar systematic agency lead-hunting), follow this specialized workflow:

Discovery and Screening

1. Read existing workspace artifacts first

   • Check Xpose Market/*.md run logs and prospect files.
   • Use the same format and directory structure to ensure findings persist across sessions.
   • If a prospect has already been screened, update the existing record instead of re-screening.

2. Live Google Maps discovery

   • Start from Google Maps listings for target geographies and service categories.
   • Capture candidate URLs, review counts, and listing evidence into the run log as you discover them.

3. Parallel candidate screening with background_spawn

   • Use background_spawn to run independent website-screening agents in parallel with Prom's main Maps discovery.
   • Each background_spawn agent must:
     • Open its own separate browser session or tab (not share Prom's Maps session).
     • Open the candidate website in that independent browser.
     • Take browser_snapshot() and browser_vision_screenshot() for visual assessment (not just text fetch).
     • Scroll/inspect the site to collect evidence (CTA clarity, design quality, contact info, service clarity, trust signals).
     • Return a structured summary or patch back to the workspace artifact with visual findings, score, and outreach angle.
   • Do not create durable subagent tasks or use deploy_analysis_team for screening unless the user explicitly asks for deeper analysis.
   • Each background agent should be independent: one agent per candidate batch, separate browser tab, concurrent execution.

4. Main Prom focus

   • Keep Prom's main browser focused on discovery (Google Maps, listing accumulation, quick qualification notes).
   • While background agents screen candidates, continue scanning more maps listings.
   • This parallelism keeps discovery moving without bottlenecking on website assessment.

Failure Handling

If background_spawn hits rate limits (Anthropic 429, timeout, etc.):

1. Note in the run log which candidates were blocked by the provider error.
2. Continue manual screening in Prom's main browser if time permits.
3. Mark those candidates as "awaiting background retry" or "text-screened only, visual pending."
4. Optionally retry later with a different model/provider.
5. Do not silently skip them or treat them as unscreened.

If browser visual inspection is unavailable (e.g., Playwright Chromium missing in runtime, headless shell not installed):

1. Fall back to text-only screening (web_fetch_batch for candidate URLs, web_search({ fetch_top_k }) for supporting evidence, one-off web_fetch for a single official site).
2. Explicitly mark the candidate as "text-screened only" in the evidence note.
3. Add a flag: visual_inspection: blocked or similar in the record.
4. Document the blocker reason.
5. Require a later visual follow-up pass before creating mockups or final pitches.
6. Do not treat text-only evidence as equivalent to visual screenshots when judging design quality or CTA clarity.

Lead Hunt → Pitch Package Follow-Through

After ranking all candidates and identifying A-tier leads:

1. Choose the top A-tier lead for the first pitch package (usually the highest-review, weakest-website candidate).

   • Example from Xpose runs: Castillo Landscaping Services (4.9 / 220 reviews, very weak visual site).

2. Create a dedicated pitch-package artifact in the workspace with:

   • Critique section: Current site weaknesses (design, CTAs, trust signals, mobile feel, conversion flow).
   • Mockup direction: What the redesigned site should include (hero headline, service cards, gallery, estimate CTA, trust badges, review proof).
   • Outreach copy: The pitch angle and value proposition (e.g., "Your reviews are doing the hard part — the site just needs to turn that trust into estimates").
   • Call script: Key talking points and differentiators for the initial outreach call.
   • Next actions: Research contact info, verify business ownership, send pitch, schedule discovery call, build mockup.

3. Link the pitch package from the run log

   • Add a line like: "Pitch Package Created: Castillo Landscaping Services"
   • This keeps the discovery work tied to outreach execution.

4. Save time on future runs

   • Once a pitch package exists, do not re-screen that candidate unless there is new evidence.
   • Link back to the package in the prospect CSV record.

---

Minimum File Outputs Per Run

1. Run log

Example:

lead-research/xpose-market/2026-04-11-frederick-run.md

Include:

• target geography
• niches scanned
• businesses reviewed
• top candidates
• skipped/disqualified reasons
• reports created

2. Master prospect sheet

Example:

lead-research/xpose-market/prospects.csv

Use one row per business. Update instead of duplicating.

3. Per-business evidence note

Example:

lead-research/xpose-market/evidence/acme-plumbing.md

Include:

• listing source and URL
• site URL
• observed issues
• score summary
• outreach angle
• status
• date checked

---

Quick Viability Heuristics

Use these as fast judgment cues during scanning:

Signal	Usually raises viability	Usually lowers viability
Website presence	No site or very weak site	Strong polished site
Conversion setup	No CTA, weak contact flow, unclear offer	Clear funnel already in place
SEO basics	Thin titles, weak structure, poor local signals	Solid location/service targeting
Business type	High-value service categories	Low-leverage categories
Outreach angle	Specific, visible fix opportunity	Vague or forced pitch

---

Anti-Patterns

• fall into blind manual scroll loops on listing pages; if scrolling is blocked, re-anchor with a real interaction or switch to bulk collection/extraction tools instead of retrying browser_scroll(...) repeatedly

Do not:

• rely only on generic web_search summaries when the user wants map-first lead hunting
• treat every ugly site as a good lead automatically
• run deep analysis on every business before screening
• keep findings only in chat without writing them to files
• confuse a cool-looking business with a high-probability client
• skip the outreach angle; weak angle usually means weak lead

---

Mental Trigger Tests

Clear match

User says: "Go through Google Maps in Frederick and find local businesses with weak websites." Expected: This skill should load.

Edge case

User says: "Find local businesses in Frederick and tell me which ones are worth outreach." Expected: This skill should load.

False positive

User says: "Explain local SEO for restaurants." Expected: This skill should not be the primary skill unless lead-finding is also requested.

---

Completion Standard

A lead-hunting run is complete when:

• businesses were sourced from live listings or equivalent local sources
• findings were persisted to workspace files during the run
• strong candidates were separated from weak ones with clear reasoning
• promising leads include enough evidence to support future outreach
• deeper reports, if created, are linked back into the saved research artifacts

---

Changelog

Date	Change
2026-04-29	v2.2.0: Added Xpose Market workflow section with background_spawn parallel-screening guidance, independent browser-session requirements for candidate evaluation, 429/timeout fallback handling, text-screened-vs-visual-screened status tracking, and lead-hunt-to-pitch-package follow-through for A-tier leads. Preserved general map-first local-lead guidance for non-Xpose work.
2026-04-23	v2.1.0: Added browser scroll-guard guidance for live listing workflows. Documented that repeated blind browser_scroll(...) retries can be blocked by the automation layer, so the workflow should re-anchor with a real interaction or switch to bulk collection/extraction patterns instead.