Most teams have three people managing tasks that one skill handles better. You give it your ICP. It researches each prospect, finds a personal angle, writes the outreach, sends, follows up, and logs everything to your CRM. You come in when someone replies. Setup under an hour.
The Problem
Most lead gen falls apart at the follow-up. Someone researches a list, writes a batch of messages, sends them, and then life gets in the way. Day 3 passes. Day 7 passes. The prospect never heard back.
The other place it breaks: the angle. Generic messages get ignored. A real opener — one specific fact about the company — makes the difference between a reply and a delete. Finding that fact for every prospect takes time most teams don't have.
The third break point is the CRM. After a week of outreach, you have no idea who's been contacted, who replied, who needs a follow-up, and who got a message that was frankly embarrassing. The record keeping is always the last thing, so it never happens.
This skill fixes all three.
The Skill
Copy the block below and paste it into Claude Code as a skill file, or paste it directly into a Claude chat. Fill in the three placeholders before you run it.
Copy The AI Lead Gen Skill
Copy this.
You are my lead generation agent. Run the full pipeline below using my ICP, CRM, and outreach channel. My ICP: [YOUR_ICP] Example: "Series A SaaS founders in the US, 10-50 employees, using HubSpot" My CRM: [YOUR_CRM_TOOL] Example: "Notion" My outreach channel: [YOUR_OUTREACH_CHANNEL] Example: "Email via SmartLead" --- Step 1 — Parse the ICP Break my ICP into searchable filters: industry, company size, geography, tech stack, and the decision-maker title I want to reach. Step 2 — Research prospects Search Apollo (apollo.io) or Clay using those filters. Pull contact names, verified emails, and LinkedIn profiles. Target 20-50 prospects per run. Stop when you hit the rate limit or the list is exhausted. Do not pad the list with low-confidence matches. Step 3 — Find a personal angle For each prospect, spend 30-60 seconds checking their LinkedIn or company news. Find one real fact: a recent funding round, a new hire, a published post, a job opening that signals growth. One genuine detail beats ten generic lines. If no angle is findable, mark the prospect as low-context and write a fallback opener based on their role and industry only. Step 4 — Write the message 3-5 sentences per message. Open with the angle you found. One sentence on what I do and who I help. One specific ask (a 20-minute call, not a demo). No "I came across your company." No feature list. No "I hope this finds you well." Template shape: Subject: [Company name] — [specific hook] Hi [First name], [Angle: one real fact about them.] [What I do, for who, with one outcome.] [Single ask — one sentence.] [My name] Step 5 — Queue the messages Add each message to [YOUR_OUTREACH_CHANNEL]. Tag each message with the prospect's ID so replies link back to the right record. Step 6 — Follow up For non-replies: send follow-up 1 on day 3 with a different angle. Send follow-up 2 on day 7 with a permission-to-close line ("Happy to drop this if the timing is off"). Stop after two follow-ups. Do not send a third. Step 7 — Log to CRM Add one row per prospect to [YOUR_CRM_TOOL] with: name, company, email, LinkedIn URL, status (contacted / replied / booked / no reply), first contact date, message sent, notes. Update the row when the status changes. --- If no prospects are returned from Apollo: - Loosen one filter at a time. Remove geography first, then company size. - Check your Apollo credit balance. Free tier: 50 enrichment credits per month. If the CRM write fails: - Confirm the API key has write scope, not read-only. - Check that field names in the CRM match exactly what I've defined above.
How to Use It
In Claude Code, put this file in ~/.claude/skills/ai-lead-gen/ and name it SKILL.md. Then open a new session and say: "Run the AI lead gen skill. My ICP is [your ICP here]. Start finding prospects."
If you use Claude on the web or desktop app, paste the block directly into a chat and fill in the three placeholders first. The agent walks you through each step.
The first run takes the longest. After that, running it weekly on a fresh batch takes about five minutes of your time.
What You Get Back
A researched prospect list — names, verified emails, LinkedIn URLs, one angle per person. Not a raw export. A worked list.
A queued message for each prospect — written with the specific angle, your ICP framing, and a single ask. Ready to send, not a template you still have to fill in.
Two follow-up messages per prospect, already scheduled for day 3 and day 7, each with a different angle. You do not have to remember who needs a follow-up.
A CRM row per prospect — status, date, message sent, and a notes field that updates when someone replies or books a call.
A clear pipeline view — who was contacted, who replied, who needs follow-up, who is done. One table, no separate spreadsheets.
What This Skill Needs
| Tool | What it's for | Cost to start |
|---|---|---|
| Apollo or Clay | Prospect research and contact enrichment | Apollo free: 50 contacts/month |
| HubSpot, Notion, or Airtable | CRM (existing one works fine) | Free tiers available |
| SmartLead or Instantly | Outreach sending and reply tracking | SmartLead: $39/mo |
| Claude Code | Agent runtime with API tool access | Max plan or API |
Apollo's free tier is enough to test the system. Most clients upgrade to Apollo Basic ($49/mo) when they hit the 50-contact ceiling and want to run it weekly.
You do not need Slack or Teams for this skill. The transcript mentioned an integration pattern where the agent lives inside messaging tools — that setup is possible but optional. The skill runs fine from Claude Code alone.
The Mistake That Makes It Fail
A vague ICP produces a useless list.
"Series A startup" is not an ICP. It gives the agent no useful filters. The result is 50 contacts who have nothing in common, no shared angle, and no reason to reply to the same message.
Here is the difference:
Weak: "US-based SaaS company, Series A, tech sector."
Strong: "Series A SaaS companies in the US with 10-50 employees where the CEO also runs sales. Revenue $1M-$5M."
The strong version tells the agent which role to search for (CEO who runs sales), which size to filter by, and what revenue signal to look for. The skill writes better angles because it can find relevant facts. The reply rate is higher because the message actually fits the person.
The formula: role + company type + geography + size + one behavioral signal. One signal. Not five.
Honest Limits
The skill does not close deals. It books conversations. What happens on those calls is yours to handle.
It does not work without an Apollo or Clay API key. The free tier covers 50 contacts per month. That is enough to run a real test, not enough to run weekly at scale.
Quality drops for low-context prospects. When there is no recent news, no LinkedIn activity, and no job postings, the fallback opener is weaker. The skill flags these. You can skip them or write a manual opener.
The follow-up sequence stops at two emails. Sending more is your decision, not the skill's. Most replies come in the first two touches. After day 7 with no reply, move on.
Install it this week with one ICP you already know. Run it on 20 prospects. Check the reply rate at day 10. That number tells you whether the ICP is right, not whether the skill works.