What is a Claude cold email skill?

Quick Answer: A Claude cold email skill is a reusable instruction set that takes prospect data as input and returns a personalised email opener, ready to send. You build it once inside Claude's Projects or as a custom skill, then run it against any lead list to generate outreach that reads like it was written by hand.

Cold email fails for one reason: it sounds like cold email. Generic openers, copy-paste value props, and zero proof you know anything about the person on the other end. A Claude skill fixes that by turning raw prospect data into a personalised first line every time, at scale, without you writing a word from scratch.

By the end of this guide, you will have a working Claude cold email personalisation skill that takes a prospect's name, company, role, and one signal (a LinkedIn post, a job change, a funding round) and returns a polished, ready-to-send opener.

What Is a Claude Cold Email Skill?

A Claude cold email skill is a saved, reusable prompt structure that lives inside Claude's Projects feature or a custom skill definition. It gives Claude a fixed role, a consistent set of rules, and a repeatable input format. Every time you feed it prospect data, it applies the same logic and returns structured output.

Think of it as a specialist on your team whose only job is to read a prospect's context and write the first two sentences of an email. They never get tired, never go off-brand, and never send a generic "I hope this email finds you well."

Why this matters for B2B SaaS teams:

• Personalisation at the opener level is the single biggest driver of reply rate improvement
• Sales reps spend 20-30 minutes per prospect on manual research and writing
• A Claude skill compresses that to under 60 seconds per lead

What You Need Before You Start

Before building the skill, gather the following:

• A Claude account with Projects access (Claude Pro or higher, or API access)
• A prospect data source: a CSV, CRM export, or LinkedIn Sales Navigator list with at least name, company, role, and one signal field
• Your email framework: the value prop, CTA, and tone you want the skill to match
• 3-5 example openers you have written manually that you consider high-quality. These train the skill's style.

Step 1: Define the Skill's Role and Rules

Open a new Claude Project. In the Project Instructions (the system prompt), define exactly what Claude is doing and what constraints apply.

Here is a working template you can adapt:

You are a B2B cold email personalisation specialist.

Your job is to write a personalised opening line (2 sentences maximum) for a cold outreach email.

Rules:
- Reference the specific signal provided. Do not invent details.
- Write in a conversational, direct tone. No corporate language.
- Never open with "I", "We", or the prospect's name.
- Do not include a value prop or CTA. That comes later in the email.
- Output only the opener. No subject line, no greeting, no sign-off.
- Maximum 40 words.

These rules do the heavy lifting. They stop Claude from drifting into generic territory and keep the output drop-in ready.

Step 2: Build the Input Format

Consistency in what you feed the skill determines consistency in what you get back. Define a fixed input structure and use it every time.

Input template:

Prospect name: [First name]
Company: [Company name]
Role: [Job title]
Signal: [One specific, recent, factual detail — a post they wrote, a company milestone, a product launch, a hire, a funding round]
Your product: [One sentence on what you sell and who it helps]

Example input:

Prospect name: Sarah
Company: Patchwork Health
Role: VP of Revenue
Signal: Patchwork just announced a Series B and Sarah posted about scaling their sales team from 4 to 20 reps this quarter
Your product: A sales enablement platform that helps high-growth SaaS teams onboard new reps 40% faster

Step 3: Run the Skill and Review the Output

Paste the input into the Project and run it. A well-configured skill should return something like:

"Scaling from 4 to 20 reps in a single quarter is the kind of growth that breaks onboarding processes before anyone notices. Curious whether you're seeing that friction yet."

That opener works because it:

1. References the exact signal (the Series B hiring push)
2. Connects it to a pain point without stating the pain directly
3. Ends with a question that invites a reply without asking for a meeting

If the output is too generic, check your signal field. Weak signals produce weak openers. "Works in SaaS" is not a signal. "Just published a post arguing that PLG is dead for enterprise deals" is a signal.

Step 4: Add Style Examples to Sharpen the Output

Claude learns from examples faster than it learns from rules alone. Add 3-5 of your best manually written openers to the Project Instructions as a style reference.

Add this block after your rules:

Style examples (match this tone and structure):

Example 1:
"Saw your breakdown of the Q3 pipeline review on LinkedIn — the point about late-stage deal velocity resonated. That's exactly the problem we built [Product] to solve."

Example 2:
"Congrats on the Shopify integration launch. Getting that live usually means the product team is finally past the infrastructure debt phase and starting to look at growth tooling."

Example 3:
"Your post about cutting CAC by 30% while growing headcount caught my attention — most teams have to pick one. Would love to know how you structured that."

Three examples is enough. Five is better. More than ten and you risk Claude averaging them rather than learning from them.

Step 5: Scale It Across a Lead List

Once the skill produces clean output on 5-10 test prospects, you are ready to run it at volume.

Option A: Manual batching Paste 5-10 formatted inputs in a single message, separated by "---". Claude will process each one and return numbered outputs. This works well for lists under 50 prospects.

Option B: API + spreadsheet Use the Claude API with a Google Sheets or Airtable integration (via Make or Zapier). Set up a trigger that reads each row, formats the input template, calls the API, and writes the output back to a new column. This scales to hundreds of prospects with no manual effort.

Option C: Claude Code If you or your team can run Claude Code, you can build a script that reads a CSV, processes each row through the skill, and outputs a new CSV with the personalised opener appended. This is the fastest option for lists over 200.

How to Evaluate Output Quality

Do not just run the skill and send. Build a quick QA step into your workflow.

Check each opener against these four criteria:

• Specific: Does it reference the actual signal, not a paraphrase of it?
• Relevant: Does the connection between the signal and your product make logical sense?
• Conversational: Would a real person write this, or does it sound like a template?
• Clean: Is it under 40 words with no filler phrases?

Flag any output that fails two or more criteria and review the input. Nine times out of ten, the signal was too weak or too vague.

Common Mistakes That Kill Output Quality

Using job title as the signal "Sarah is a VP of Revenue" is not a signal. It is a data point. Signals are events, actions, or statements. Find something that happened recently.

Overloading the input More data does not mean better output. One strong signal beats five weak ones. If you give Claude five signals, it will try to reference all of them and produce a cluttered opener.

Skipping the style examples Without examples, Claude defaults to its own interpretation of "conversational B2B tone," which is usually fine but not yours. Examples anchor the output to your voice.

Not testing on real prospects first Run the skill against 10 prospects you know well before sending to anyone. You will catch edge cases in your rules before they reach an inbox.

FAQs

What is a Claude cold email skill and how does it work? A Claude cold email skill is a reusable prompt configuration saved inside Claude's Projects feature. It gives Claude a fixed role, rules, and an input format. You feed it prospect data (name, company, role, and a signal), and it returns a personalised email opener. The skill applies the same logic every time, so output quality stays consistent across hundreds of prospects.

How is this different from using ChatGPT or a generic AI prompt for cold email? The difference is in the saved structure. A Claude skill stores your rules, tone guidelines, and style examples permanently inside a Project. You do not re-explain your requirements each session. ChatGPT and one-off prompts require you to rebuild context every time, which leads to inconsistent output. Claude's Project system keeps the skill intact and ready to run.

What kind of signals produce the best personalised openers? The best signals are recent, specific, and public. LinkedIn posts, company announcements, funding rounds, product launches, new hires, and published content all work well. Avoid signals that are too generic (industry, job title) or too obscure (signals the prospect would not recognise as relevant to them). One strong signal outperforms three weak ones every time.

Can I use this skill with a cold email platform like Instantly or Salesforge? Yes. The skill outputs plain text openers that you can paste directly into any email platform. For higher-volume use, connect the Claude API to your platform via Make or Zapier. Some cold email tools are also building native Claude integrations, but a direct API connection gives you more control over the input format and output quality.

Is a Claude Pro account enough to build this skill, or do I need the API? Claude Pro is enough to build and test the skill manually inside Projects. For automated, high-volume use across a lead list, you need API access. The manual approach works well for teams sending fewer than 50 personalised emails per week. Beyond that, an API-based workflow saves significant time.