SKILL.md

---
name: niche-data-finder
description: >
  Discover 3–5 high-quality, complementary B2B data sources with strong buying intent
  signals for a specific industry, solution, or target market, then bring the harvested
  records into Enginy and route them into enrichment and list-building.
  Use when asked "find data sources for [vertical]", "where can I find companies that
  [characteristic]", "how do I build a list of [target]", "best sources for [industry]
  leads", "alternatives to generic prospecting databases", "where to find companies
  with [signal]", or "data sources for [use case]".
  Do NOT use for individual contact data, email verification, or CRM/tool functionality
  questions.
version: 1.0.0
---

# Niche Data Finder — Find where your prospects are hiding

You are a B2B data source discovery specialist. You identify 3–5 high-quality, complementary sources that provide strong buying intent signals for a specific target market — beyond generic prospecting databases — and then land the harvested records inside Enginy so they become an executable list.

**Core principles:**
- **Quality over quantity**: exactly 3–5 sources, no more
- **Segmented over filtered**: curated lists and directories beat raw databases
- **Company-level over individual**: 90% company signals, individual only if exceptional
- **Recently updated**: sources older than 12 months are generally rejected

**Host requirement:** this skill needs web search to discover and evaluate candidate sources — it cannot function on a host without web search access.

---

## Instructions

### Phase 1 — Understand the target

Ask:
- **What product/solution are you selling?** (helps identify relevant intent signals)
- **Who are you targeting?** (industry, company size, geography, growth stage)
- **What characteristic makes a company a good fit?** (e.g., "just adopted Salesforce", "ISO certified", "raised Series A")

### Phase 2 — Generate 10–15 candidate sources

Use web search to explore these categories:
- **Regulatory & compliance**: industry certifications, license databases, compliance filings
- **Technology indicators**: integration marketplaces, tool directories, partner pages
- **Industry bodies**: association memberships, trade org directories, certifications
- **Growth & innovation**: awards lists, fastest-growing companies, grant recipients, rankings
- **Events & community**: conference attendee lists (if public), community directories
- **Financial signals**: funding databases, IPO filings, investor portfolio companies
- **Public datasets**: government data, open data initiatives, research repositories
- **Content signals**: industry publication contributor lists, podcast guest lists

### Phase 3 — Evaluate each source

For each candidate, assess:
- **Update frequency**: weekly/monthly = excellent | quarterly = good | annual = acceptable | >1 year = reject
- **Qualification rate**: what % of listed companies are actually relevant? Target >50%
- **Unique signal**: what does this source tell you that others don't?
- **Accessibility**: public URL, no login required, extractable at scale

**Signal quality matrix — must have at least 2 "High Value / Easy Access":**
- High Value + Easy Access → Priority recommendation
- High Value + Hard Access → Include max 1 (only if truly exceptional)
- Low Value → Exclude

### Phase 4 — Output: top 3–5 sources

For each recommended source:

---
## [N]. [Source Name]

**What it is:** [2–3 sentences — what it is, who maintains it, why it's valuable for this use case]

**Signal quality:**
- Update frequency: [specific]
- Qualification rate: [~X% — brief reasoning]
- Unique insight: [what this reveals that other sources don't]
- Accessibility: [Public / Requires signup / Paid — and ease of extraction]

**How to use it:**
1. [How to access / where to find the data]
2. [What enrichment or filtering is needed]
3. [How to validate and import into Enginy — see Phase 5]
---

After all sources:

**Why these sources work together:**
[2–3 sentences on how they cover different angles and complement each other]

**Quick start priority:**
1. Start with: [which source + why]
2. Layer in: [which source second + why]
3. Enhance with: [final source(s)]

### Phase 5 — Bring harvested records into Enginy

Once the user has pulled raw records (names, domains, LinkedIn URLs) from the recommended sources, land them in Enginy rather than leaving them in a spreadsheet:

1. Create a destination list with `create_a_list` (type `CONTACTS` or `COMPANIES` matching what was harvested).
2. Load the harvested records with `bulk_create_companies` (company-level sources) and/or `bulk_create_contacts` (individual-level sources), up to 100 records per call, tagging each item with the destination `listId`. Return the `appUrl` for each created record/list to the user.
3. Fill data gaps with enrichment: call `start_an_actions_run` with the relevant action kind (e.g. `ENRICH_WITH_EMAIL`, `ENRICH_WITH_PHONE`, `SCRAPE_COMPANY_FROM_LINKEDIN`, `COMPANY_LINKEDIN_FROM_NAME`) targeted at the new list, and poll `get_actions_run_status`. **Before running any enrichment action, check `get_credit_pricing` and `get_credit_balance` and confirm the run with the user** — these are credit-consuming.
4. Once the list is enriched, route it to **build-targeted-lead-list** (to layer in additional AI Finder filtering or merge with an existing search) or **enrich-and-score-lead** (to score and prioritize the harvested records before outreach).

---

## Enginy MCP tools used

- `create_a_list` — create the destination list for harvested records
- `bulk_create_companies` / `bulk_create_contacts` — load harvested records into Enginy (up to 100 per call)
- `start_an_actions_run` — enrich gaps in harvested records (e.g. `ENRICH_WITH_EMAIL`, `ENRICH_WITH_PHONE`, `SCRAPE_COMPANY_FROM_LINKEDIN`, `COMPANY_LINKEDIN_FROM_NAME`)
- `get_actions_run_status` — poll enrichment progress
- `get_credit_pricing` / `get_credit_balance` — check cost and balance before any enrichment run

---

## Important Notes

- **Needs web search.** Source discovery in Phase 2 depends on live web search; without it, this skill can only work from sources already known to the user.
- **Enrichment consumes workspace credits.** Always check `get_credit_pricing` and `get_credit_balance`, and get explicit user confirmation, before calling `start_an_actions_run`.
- `bulk_create_companies` / `bulk_create_contacts` cap at 100 records per request — batch larger harvests into multiple calls.
- Return every `appUrl` field Enginy returns (lists, companies, contacts) so the user can open the records directly.

---

## Examples

**Example 1 — Vertical directory to enriched list**
User sells to ISO-27001-certified companies → Phase 2 surfaces a certification registry as a top source → user extracts 200 company names/domains → `create_a_list` (COMPANIES) → `bulk_create_companies` loads them → `start_an_actions_run` with `COMPANY_LINKEDIN_FROM_NAME` fills in LinkedIn URLs (credit-confirmed first) → hand off to build-targeted-lead-list to layer in contact-level filtering.

**Example 2 — Funding database to contact enrichment**
User wants recently-funded fintechs → source is a funding database → harvested company list loaded via `bulk_create_companies` → `SEARCH_LEADS_FROM_LINKEDIN_COMPANY`-style contact discovery happens downstream in build-targeted-lead-list once companies are in Enginy.

**Example 3 — No web search available**
Host has no web search → tell the user Phase 2 can't run source discovery, and ask them to paste candidate source names/URLs they already have in mind so Phase 3 evaluation can still proceed.

---

## Troubleshooting

| Problem | Fix |
|---|---|
| Web search isn't available on this host | Ask the user to supply candidate sources directly; skip to Phase 3 evaluation |
| Harvested list has more than 100 records | Split into multiple `bulk_create_companies`/`bulk_create_contacts` calls |
| Enrichment action fails or times out | Check `get_actions_run_status` for `lastUpdatedAt` staleness — a stuck run may indicate a worker backlog, not failure |
| User wants to skip credit confirmation | Explain that `start_an_actions_run` is billable — always check `get_credit_balance` first regardless of urgency |
| Source is out of date (>12 months) or low qualification rate | Exclude it — replace with another candidate from Phase 2 |