recruiting-sourcer

Recruiting Sourcer

Overview

This skill produces a tiered candidate shortlist for one role: people whose current title and seniority fit, at the companies and in the locations the search asked for, with people at the hiring company and do-not-poach accounts removed, each with a validated email where one exists and a mobile for the top tier. The report shows the search strings, the counts, the exclusions, the contact coverage, the cost and the outreach plan.

It prevents four sourcing mistakes:

1. One title string. "Senior React Engineer" misses "Frontend Engineer II", "UI Engineer" and "Software Engineer, Web". Titles expand into a family with exclusions, as sourcers have done in Boolean strings for years.
2. Sending the hiring manager's own team a recruiting email. The Lead Database has no "not this company" filter, so the skill removes the client's domains, subsidiaries and do-not-poach accounts after the pull and reports the count.
3. Paying for mobiles on 300 people when the recruiter will call 40. Mobile Finder ($0.25 per number found) runs on the A tier only.
4. One source tag for every candidate. The agency cannot tell which search produced the replies and hires, so the next search repeats the weak one. Each search gets its own source in the ATS import.

When to Use

• "Source senior React engineers in Austin and give me their emails and mobiles."
• "Find VPs of Engineering at Series B fintechs on the East Coast."
• "Map the account executives at these 20 competitors."
• "Here's my Apollo search for data engineers; turn it into a contactable shortlist."
• "Build a pipeline of controllers for a client in Chicago."

Use a different skill when:

Prerequisites

Authoritative Foundations

Details and routing: references/framework-notes.md.

Workflow

1. Write the role brief

Fill the Role Brief table: title as written, title family, seniority, location, pay band, two selling points, target companies, company traits, must-haves that the data can filter (title and level) and must-haves it cannot (skills, years of experience, visa status, and for sales roles quota attainment and deal size; those are screened by a human). List the excluded employer domains.

Build the target-company list by reason and tag each company: ahead (one or two stages or size bands ahead of the client), competitor, alumni (the hiring team's past employers), adjacent (same buyer, different product). Reasons and why each replies: references/candidate-messages-and-handoff.md section 1.

2. Expand the title into a family

From references/title-expansion-library.md, build:

• title: comma-separated synonyms. The Lead Database matches each term partially, so "Frontend" also matches "Senior Frontend Engineer".
• notTitle: terms that partial matches drag in (Native for a web React role, Manager and Director for an individual contributor role, Recruiter, Intern, Instructor, Contract).
• seniority and department arrays. The tool documents vp, director and c_suite as seniority values and sales and engineering as departments; probe any other value with limit: 1 and read the returned rows before relying on it.

3. Pick the source and count

Run two searches when there is a poach list: one with companyDomain (target companies, comma-separated) and one market-wide with location and company traits. Count both. For engineering roles where the market-wide count is thin, add a GitHub user search (location:, language:, repos:) as a free second source and match profiles back by name and company (references/sourcing-runbook.md section 9).

4. Pull

Database: page with after: "0" then pagination.next_after, limit: 100, until the planned rows. Apollo: register the webhook at app.scrapercity.com/dashboard/webhooks, start the run, save the runId, and stop the session; the webhook (or a check every few minutes) tells you when it reaches SUCCEEDED. Runs usually take 15 to 30 minutes.

5. Exclude, dedupe, QA

1. Union the searches; dedupe on LinkedIn URL, then email, then first name + last name + company domain.
2. Remove the hiring company: every row whose company_domain is a client domain or subsidiary. Report the count.
3. Remove do-not-poach accounts (the agency's current clients and partners).
4. Remove titles that slipped through: read a random sample (the size is the recruiter's call; the example reads 20) and record title fit, level fit and location as "N of M", plus a few LinkedIn URLs opened to confirm the person. When misfits share a pattern (one drag-in title, one wrong level), fix notTitle or the company filters and re-pull rather than hand-deleting.
5. Remove people the client already has in its ATS or rejected within the client's look-back window, if the client supplies the list.

6. Tier

Tier each candidate by the evidence the row holds, and write it to fit:

Location, company size and keywords are filters, so every pulled row already meets them.

7. Add contact data

1. Validate Lead Database emails with validate_emails.
2. For A and B candidates with no email, find_emails with {first_name, last_name, domain} ($0.05 per email found).
3. For A candidates only, find_mobiles with the LinkedIn URL, else the work email ($0.25 per number found). One input per person.
4. Normalize mobiles to E.164.

8. Draft the messages, hand off, read the replies

1. Write the first email (under 400 characters: one specific line about the candidate's work, the role in one line, the pay band and setup, one selling point, a 15-minute ask), a follow-up after day 7, a LinkedIn connection note, and a closing email that asks for a referral. Send Sunday to Thursday. The LinkedIn note and the close go out on days the recruiter sets. Templates and the data behind each rule: references/candidate-messages-and-handoff.md sections 2 and 3.
2. Import the shortlist into the ATS as prospects with the source set per search (target companies by reason, market-wide), so replies and hires count per search. Greenhouse mapping is in section 5.
3. Tag replies interested, not_now, not_looking, wrong_fit, referred, opt_out. An opt_out goes on the agency suppression list the same day, every channel. After the first wave, compare wrong_fit share by title term and reply rate by target reason: the title term that draws the most wrong_fit moves to notTitle, and the next pull shifts toward the reasons that reply (section 4).
4. Before sending a shortlist that has waited, re-check title and employer; set how long it may wait from the share of candidates who had moved at your last re-check. Write the shortlist CSV, the long list CSV and the report from templates/output-template.md; run scripts/check-output.py.

ScraperCity Implementation

Tools and prices (from references/scrapercity-catalog.md)

Calls

{"tool": "query_lead_database", "args": {
  "title": "React, Frontend, Front End, Front-End, UI Engineer, JavaScript Engineer, Web Engineer",
  "notTitle": "Native, Manager, Director, Head of, VP, Recruiter, Intern, Instructor, Designer, QA, Contract",
  "seniority": ["senior"],
  "city": "Austin", "state": ["Texas"], "country": ["United States"],
  "keywords": ["fintech", "saas"],
  "hasEmail": false, "limit": 1}}
{"tool": "query_lead_database", "args": {
  "companyDomain": "pinecrestpay.example,tallyworks.example,orbitfin.example",
  "title": "React, Frontend, Front End, Front-End, UI Engineer, JavaScript Engineer",
  "notTitle": "Native, Manager, Director, Head of, VP, Recruiter, Intern", "limit": 100, "after": "0"}}
{"tool": "query_lead_database", "args": {"url": "PASTE_APOLLO_PEOPLE_SEARCH_URL", "limit": 1}}
{"tool": "scrape_apollo", "args": {"url": "PASTE_APOLLO_PEOPLE_SEARCH_URL", "count": 1000, "fileName": "data-eng-chicago-20261013"}}
{"tool": "find_emails", "args": {"contacts": [{"first_name": "Maya", "last_name": "Chen", "domain": "tallyworks.example"}]}}
{"tool": "find_mobiles", "args": {"inputs": ["https://www.linkedin.com/in/maya-chen-frontend"]}}

hasEmail: false on the count keeps candidates without an email in the pool; Email Finder fills them later. Set it to true only when the budget rules out Email Finder. Replace each PASTE value with the URL copied from the browser.

Apollo through the CLI: npx scrapercity apollo "<url>" with a count; the CLI prints that Apollo usually takes 15 to 30 minutes and points to the webhook page. Apollo URLs must use Filters 1.0 format per the catalog.

Worked cost estimate

Senior React engineers in Austin for a fintech client, Growth plan, shortlist of 320 (A 84, B 236).

Ceiling before the run: $0.42 + 69 x $0.05 ($3.45) + 84 x $0.25 ($21.00) = $24.87. Show it and wait for a go-ahead. If the user were on Starter and used an Apollo URL instead, a 1,000-lead run would add 1,000 x $0.0039 = $3.90 and a 15 to 30 minute wait.

Errors

Output Format

1. shortlist-[role]-[date].csv: A and B candidates with first_name, last_name, title, seniority, company_name, company_domain, person_city, person_state, linkedin_url, tier, fit, email, email_status, mobile (A tier, E.164), source, pulled_at.
2. longlist-[role]-[date].csv: C tier, no paid enrichment.
3. Report from templates/output-template.md: role brief with pay band and selling points, search strings, count check with the QA line, sourcing funnel, contact coverage, cost, shortlist sample, outreach plan, message drafts, next steps with the ATS import and reply tags.

What the reader receives is the deliverable only: product names (Lead Database, Apollo Scraper, Email Finder, Email Validator, Mobile Finder), never tool ids, internal file paths or reference section numbers; filter parameter names appear only in the Search Strings table; another skill is named only in a handoff line; no tables of checks that were not run (one "Not checked:" line instead); no repeated identical rows.

Quality Check

• The title family and notTitle set are written out, and the QA line reports title fit, level fit and location as "N of M".
• No candidate's company domain matches the hiring company, a subsidiary or a do-not-poach account, and the removed count is reported.
• Every shortlisted row's tier matches its fit evidence.
• Apollo runs, if any, had a webhook registered and a count between 500 and 50,000.
• Every email in the shortlist was validated or came from Email Finder; catch-all emails are marked.
• Mobiles exist only for A-tier rows and are E.164.
• The Role Brief states a pay band in dollars, and the first message carries it.
• Message Drafts holds at least the first email and the closing email, each under 400 characters, with no bracketed placeholders left for the candidate to see.
• Next Steps names the ATS import with the source per search and the six reply tags.
• scripts/check-output.py report.md prints PASS.

Common Pitfalls

Execution Artifacts

Related Skills

• icp-list-builder and decision-maker-finder for the same filters aimed at buyers.
• mobile-appender for mobiles on an existing list of profiles.
• email-list-cleaner to re-validate a candidate list before a new search.
• cold-email-copywriting for candidate outreach copy.
• multi-channel-outreach to combine email, calls and LinkedIn.
• scrapercity-lead-database for every Lead Database filter in depth.