A Claude skill that writes the cold email itself
The whole email, written to the limits, with the frameworks stripped of invented results.
Nothing gated, no email address, MIT licensed, adapted from open-source work and credited below
Why this one exists
The judgements the file encodes
This is the part a general prompt does not give you. Each one is a decision we have already made and written down.
| It encodes | Which means |
|---|---|
| The frameworks, with the fake results removed | The upstream examples carried named companies achieving specific percentages, none of which we can verify and none of which belong in a file people copy. The structures are kept, because they are genuinely useful thinking patterns, and the specifics are left blank with an instruction to fill them with things that are true. |
| Four levels of personalisation, and which one a campaign can hold | A merged first name, a role, something the company did, something that person wrote. The third is the working floor for a campaign and the fourth is for the list you would ring by hand. Naming the levels stops the argument about whether a campaign is personalised: it is, at a level, and the level is a capacity decision. |
| The limits that protect the domain | Ninety-nine words, a sixteen-word opening, no links in a first email, no meeting ask in a first touch, and outbound sent from a domain bought for it rather than the one the business invoices from. Those are ours. The last one is the difference between an inconvenience and a week nobody enjoys. |
| Subject-line data, with its publishers named | The figures on short lower-case subject lines are real and useful and belong to email tooling companies who sell email tooling. They sit in one section that says so, with the caution that open rates depend on a tracking pixel which is itself a deliverability cost, so every open-rate figure is softer than it looks. |
| Sequence length as arithmetic, not preference | Before choosing five steps, the file states what five costs: every person consumes one send per step, so the daily new-lead rate is capacity divided by steps. Two extra steps take roughly a third of the people you will ever reach, so they have to earn it with a genuinely new angle each. |
Provenance, and what we changed
This file started as somebody else's open-source work, under a licence that lets us adapt and republish it. Here is the whole of what that means for this one.
| Field | Value |
|---|---|
| Based on | cold-email-writer in coreyhaines31/marketingskills, pinned to the commit we read it at. |
| Original author | Corey Haines, under the MIT licence. |
| Adapted by | TheCold.Ai, on 2026-09-18. |
| What changed |
|
| This file's licence | MIT. Copy it, change it, use it with your own clients. |
The same information is in the frontmatter of the file itself and in a Credits section at the bottom of it, so it stays attached when somebody copies the file onto their own machine. The licence evidence, including the date it was read, is kept in the repository this site is built from.
SKILL.md, in full
Exactly what is committed, frontmatter included. Copy it into a folder named after the skill, wherever your assistant reads them from.
---
name: cold-email-writer
description: >-
Writes B2B cold emails and short sequences that read as though a person wrote
them. Use when someone wants a cold email, a prospecting or outbound email, an
SDR email, a campaign, or says nobody is replying to their emails. Covers
subject lines, opening lines, the body, the ask, personalisation and follow
ups. Use before sending anything cold, and before rewriting a campaign that is
getting no answers.
license: MIT
metadata:
based_on: https://github.com/coreyhaines31/marketingskills/blob/5b2c0007766c6a1cf1d53fd8fc73e979e0821022/skills/cold-email/SKILL.md
upstream_repo: coreyhaines31/marketingskills
upstream_author: Corey Haines
upstream_license: MIT
adapted_by: TheCold.Ai
adapted_on: "2026-09-18"
changes:
- British English throughout; dashes and emoji removed
- Folded into one file from a skill plus five reference files, dropping the relative links
- Every invented customer result in the framework examples removed; the frameworks kept as structures
- Published subject-line figures moved into one attributed section with their publishers named
- Our own limits added: ninety-nine words, no links in a first email, no meeting ask in a first touch
---
# Cold email writer
You write cold emails that sound like a sharp, thoughtful person wrote them, not
a machine filling in a template.
## Before writing
Understand the situation. Ask for whatever is missing, in one message, and work
with what you are given rather than blocking on a full set.
1. **Who are you writing to?** Role, company, and why them specifically.
2. **What do you want?** A reply, an introduction, a meeting later.
3. **What is the value?** The specific problem you solve for people like them.
4. **What is your proof?** A result you can name, or nothing. Never invent one.
5. **Any signals?** Funding, hiring, a post, news, a change of tools.
If there is a real signal and a clear value, that is enough to write. Say what
would have made it stronger rather than refusing to start.
## Principles
### Write as a peer, not a vendor
It should read as though it came from somebody who understands their world. Use
contractions. Read it aloud. If it sounds like marketing, rewrite it.
### Every sentence earns its place
Cold email is ruthlessly short. If a sentence does not move the reader towards
replying, cut it. The best ones feel as though they could have been shorter.
### Personalisation has to connect to the problem
Remove the personalised opening. If the email still makes sense, the
personalisation was decoration. The observation should lead into why you are
writing.
Four levels, worst to best:
1. Their name and company merged into a template. Everybody can tell.
2. Their role or industry. Better, still generic to a thousand people.
3. Something the company did: hiring, a launch, a new site, a change.
4. Something specific to that person: what they wrote, said, or built.
Level 3 is the working floor for a campaign. Level 4 is for the list you would
call by hand.
### Lead with their world
They should see their own situation reflected back. "You" and "your" should
outnumber "I" and "we". Do not open with who you are.
### One ask, low friction
An interest-based ask beats a meeting request. One per email. Make it possible
to say yes in one line.
## Voice
A smart colleague who noticed something relevant and is passing it on.
Conversational, not sloppy. Confident, not pushy.
- **Senior**: shorter, understated, peer level.
- **Middle**: more specific about the work, slightly more detail.
- **Technical**: precise, no padding, no explaining their own job to them.
It should not sound like a template with fields swapped in, a pitch deck
compressed into a paragraph, or a message generated by something that has read a
lot of cold emails: no "I hope this finds you well", no "I came across your
profile", no leverage, synergy or best-in-class.
## The limits
These are ours and they are not preferences.
- **Ninety-nine words** in the body of a first email, a hundred and fifty after.
- **Sixteen words** in the personalised opening line.
- **No links in a first email.** Not the site, not the calendar, not the
signature. This is a domain that has never written to this person; a link is
the easiest thing in the message for a filter to score, and the action you
want is a reply.
- **No meeting ask in the first touch.** The ask that belongs there gives
something away and needs one word back.
- **No invented number**, result or customer outcome, in any email, ever.
- **Send from a domain bought for outbound**, never the one the business
invoices from. A reputation problem on a sending domain is an inconvenience.
The same problem on your primary domain stops your invoices arriving.
## Structures that work
Frameworks are thinking patterns rather than templates. Pick the one that fits,
or write freely if it flows without one. The examples below are structures with
the specifics left out on purpose: fill them with things that are true.
**Problem, agitate, solution.** Name the pain, show what it costs downstream,
then what changes. The workhorse, for a reader who knows they have the problem
but has not gone looking for an answer.
**Before, after, bridge.** Where they are now, where they could be, what
connects the two. For offers with a clear change of state.
**Question, value, ask.** One sharp pain question, one line of value, one next
step. For senior readers who want it over quickly.
**Trigger, insight, ask.** Something happened, here is what it usually creates,
here is something useful. The strongest shape when a real signal exists, as long
as you do not open with congratulations.
**Story, bridge, ask.** A firm like theirs had the problem, this is what they
did. Only usable if the story is real and you are allowed to tell it. A story
with the names filed off still has to have happened.
## Subject lines
Short, boring, internal-looking. Its only job is to get the email opened.
- Two to four words, lower case, no punctuation tricks.
- It should look like something a colleague sent: "reply rates", "hiring ops",
"second site", "quote turnaround".
- No product name, no urgency, no emoji, no first name.
- A question only when it is specific to a real pain. Generic questions fail.
## Follow-ups
Each one adds something new: a different angle, a different consequence, a
useful resource. "Just checking in" gives nobody a reason to answer.
- Three to five emails in total, with widening gaps.
- Each stands alone. They may not have read the earlier ones, and reminding them
that they ignored two emails is not a strategy.
- The last email closes the loop cleanly and means it.
Before choosing five steps, know what it costs. Every person added to a campaign
consumes one send for each step, so the number of new people a fleet can take
each day is its daily capacity divided by the number of steps. A fleet sending
240 a day sustains 48 new people on five steps and 80 on three. Illustrative
figures, not measured from any account. Two extra steps cost roughly a third of
the people you will ever reach, so they have to earn it.
## Benchmarks published by others
Figures from published analyses by email tooling companies, named so you can
weigh them. They are their users and their samples, mostly outside the UK, and
we have not measured any of them. A reference range, not a target.
- Two-word subject lines are reported to get substantially more opens than five
word ones, and moving from two words to four is reported to reduce replies
(Lavender).
- Two to four words returned about 46% open rates against 34% for ten words, in
an analysis of several million emails (Belkins).
- All lower case had the highest open rates across a very large sample, and
subject lines that look internal rather than commercial roughly doubled opens
(Gong).
- A first name in the subject line is reported to reduce replies by around 12%,
while contextual personalisation, a pain, a competitor, a trigger, increases
opens.
- Reported penalties: salesy verbs, urgency words, exclamation marks, numbers
and percentages in the subject, and pitching the product there.
- Senior readers reply less often overall but respond more often than junior
ones once an email passes their filter.
Two cautions before quoting any of it. The publishers sell email tooling, and
open rates depend on a tracking pixel that is itself a deliverability cost, so
treat every open-rate figure as softer than it looks.
## Quality check
Before handing anything over:
- Does it sound like a person wrote it? Read it aloud.
- Would you reply to it?
- Does every sentence serve the reader rather than the sender?
- Does the personalisation connect to the problem, or is it decoration?
- One clear, low-friction ask?
- Under the word limit, with no links and no invented figure?
## What to avoid
- "I hope this email finds you well." "My name is X and I work at Y."
- Jargon: synergy, leverage, circle back, best-in-class, leading provider.
- Feature lists. One proof point beats ten features.
- HTML, images, or any link in a first email.
- A fake "Re:" or "Fwd:" subject line. It works once and costs you the account.
- The same template with only the first name swapped.
- Asking for thirty minutes in a first email.
- "Just checking in."
## Not for you if
- You are writing to people who opted in. This is built for the version where
nothing is owed to you, and it will read as too cautious for a warm list.
- You want volume. The limits here exist to protect the domain that makes volume
possible at all.
- You have not checked who you may lawfully email. In the UK that depends on the
legal form of the recipient rather than the address, and it is a question to
settle before the copy, not after.
---
## Credits
Adapted by TheCold.Ai, a B2B outbound agency in Middlesbrough, England, from
`cold-email` in Corey Haines's marketing skills repository
(github.com/coreyhaines31/marketingskills), published under the MIT licence,
Copyright (c) 2025 Corey Haines, as stated in the LICENSE file of that
repository. The skill and its five reference files were folded into this single
file. What changed is listed in this file's frontmatter under
`metadata.changes`. The upstream notice is reproduced in full in the LICENSE
file of the download bundle at thecold.ai/claude-skills/. Not affiliated with
Anthropic.
MIT licensed. Copy it, change it, use it with your own clients. Save it as SKILL.md inside a folder named after the skill. The description line in the frontmatter is what decides when your assistant reaches for it, so if you edit anything, edit that last.
This is not worth your time if
Questions about this skill
Straight answers. If yours isn't here, ask it on the call: we'd rather tell you no than sell you the wrong thing.
The rest of writing and replies: copywriting-refiner, copywriting-first-touch, copywriting-follow-up and reply-handler. Everything sits on the library page, next to the copy audit, which is the one people usually want after this. What we do for money is outbound, and none of it is a condition of using this.
Claude is a product of Anthropic. TheCold.Ai is not affiliated with, endorsed by or connected to Anthropic.