Leads

Lead Pipeline Skill

Full automated pipeline: scrape → ICP filter → email verify → personalize → CSV

Pipeline lives at pipeline.py. All leads export to output/.

---

Setup (one-time per machine)

git clone https://github.com/jayden3455-collab/Claude-leads-skill.git
cd Claude-leads-skill
pip3 install -r requirements.txt
cp .env.example .env
# Fill in the three required keys (see below)

Required API keys (fill into .env)

| Key | Where to get it | Purpose | |-----|----------------|---------| | ANTHROPIC_API_KEY | console.anthropic.com → API Keys | ICP filter + personalization | | AI_ARK_API_KEY | ai-ark.com → Developer Portal | Lead scraping database | | MILLION_VERIFIER_API_KEY | millionverifier.com → API | Email verification |

Never hardcode keys — always use the .env file.

---

Running the pipeline

python3 pipeline.py ""  [personalization_style]

Arguments

| Arg | Required | Description | |-----|----------|-------------| | ` | Yes | Plain-English description of who to target | | | Yes | Hard cap on leads to process. **= AI Ark credits spent.** | | [personalizationstyle] | No | casestudy (default), topservice, topcompetitor, none` |

Credit math

• Each lead processed through email export = 1 AI Ark credit
• Set max_leads to stay under your credit budget
• Safe default: 9000 (leaves buffer in a 10k-credit account)
• Typical email hit rate: 8–13% depending on niche

Example invocations

# Boutique M&A advisory firms, US, decision makers
python3 pipeline.py "Buy-side M&A advisory firms in the US, company size 5-100 employees, decision makers: Managing Partner, Managing Director, Partner, Principal, CEO, Founder" 9000

# PR agencies, broader company size
python3 pipeline.py "PR agencies and public relations firms in the US, company size 5-200 employees, decision makers: CEO, Founder, Owner, Partner, Managing Director, Director, VP, President" 9000

# Podcasting agencies
python3 pipeline.py "Podcasting agencies and podcast production companies in the US, company size 5-100 employees, decision makers: CEO, Founder, Owner, Director" 5000 case_study

# No personalization (faster, cheaper on Claude tokens)
python3 pipeline.py "SaaS companies, US, 10-200 employees, founders and CEOs" 9000 none

---

Pipeline stages

| Stage | What happens | Output | |-------|-------------|--------| | 1. Scrape | Claude converts ICP → AI Ark filters, fetches profiles + emails | output/checkpoints/01_scraped_*.csv | | 2. ICP filter | Claude Haiku scores each lead PASS/FAIL against the ICP | output/checkpoints/02_icp_filtered_*.csv | | 3. Email verify | Million Verifier bulk-verifies all emails, keeps ok + catch_all | output/checkpoints/03_verified_*.csv | | 4. Personalize | Visits each company website, extracts top case study or service | Final CSV | | 5. Export | Final CSV lands in output/pipeline_.csv | output/pipeline_*.csv |

---

Output CSV columns

First Name, Last Name, Full Name, Email, Email Status, Title, Seniority, Department, LinkedIn URL, Person Location, Person City, Person State, Person Country, Company, Company Domain, Company LinkedIn, Company Size, Company Industry, Company HQ City, Company HQ Country, ICP Match, Personalization, Top Competitor

---

How to pick max_leads and ICP

Company size guidance:

• Too small to have budget: under 5 employees
• Sweet spot for cold email clients: 5–100 employees
• Has in-house teams: 200+ employees

Niche pool sizes (AI Ark, approximate):

• Hyper-niche (M&A advisory, podcasting agencies): 3,000–5,000 people
• Mid-size niche (PR agencies): 10,000–20,000 people
• Broad (marketing agencies, SaaS): 50,000+ people

Expected yield per 9,000 credits:

• 8–13% email hit rate → 720–1,170 with emails
• ICP filter removes ~40–60% → 288–700 pass
• MV verification removes ~1–5% → 250–680 verified leads

To hit 2,000+ verified leads, either:

1. Run multiple niches and merge the CSVs
2. Target a broad niche with a large pool

---

When a run finishes

Report:

• Niche targeted
• Credits used (= max_leads capped at pool size)
• Final verified lead count
• Output file path

If the pool is smaller than max_leads, credits used = pool size (AI Ark stops at last page).

---

Gotchas

• Email hit rate is not 100% — AI Ark only has emails for 8–13% of people in most niches. This is normal. The rest are scraped as profiles but dropped before ICP filtering.
• Million Verifier can queue-delay — small batches (<200 emails) sometimes sit at 0% for 20–30 minutes before jumping to done. Don't kill the process.
• catch_all emails are kept — MV pipeline keeps both ok and catch_all statuses. Only ok emails should go into cold email campaigns. Filter by Email Status = ok before uploading to Smartlead/PlusVibe.
• AI Ark credits are spent on export, not search — browsing profile pages is free; the /people/export/single call (which fetches the email) costs 1 credit each.
• Never kill a running pipeline — credits are already spent at the scrape stage. Killing mid-run orphans leads that are in the checkpoint but won't make it to the final CSV. Always let it finish.
• Checkpoints are your safety net — if the pipeline crashes after step 2, the ICP-filtered leads are saved. You can re-run verification manually against 02_icp_filtered_*.csv.
• python3, not python — the system command is python3. python is not aliased on macOS by default.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: jayden3455-collab
• Source: jayden3455-collab/Claude-leads-skill
• License: MIT

Install and usage instructions live in the source repository linked above.