Cold Outreach skill for Claude Code

Cold outreach is done well when a small, accurately targeted list gets short messages that show a real reason for contacting this specific person, every message is lawful in the recipient’s country and easy to decline, opt-outs are honoured everywhere at once, and the programme is judged by positive replies and meetings, not sends or opens.

Ground rules

• Relevance before volume. Fifty accounts with a researched reason beat 5,000 scraped addresses; low relevance shows up as spam complaints that damage the domain for everyone.
• Personalization must be true and checkable. Never fabricate a mutual connection, a referral, prior contact (“following up on our call”), or a fake “Re:” / “Fwd:” subject.
• One ask per message, and make it small (a reply, a yes/no, a 15-minute call).
• Every message carries a clear way to stop, and a stop request ends contact on every channel and from every sender in the company.
• Do not run unauthorised automation or scraping on LinkedIn; it breaches LinkedIn’s terms and gets accounts restricted.
• Check the law for each recipient country before the first send (see the legal section).
• Do not send anything on the user’s behalf. Draft; the user reviews and sends.

Workflow

- [ ] 1. Define the ICP and the trigger that makes outreach timely
- [ ] 2. Build and verify the list; record source and jurisdiction per contact
- [ ] 3. Confirm the legal basis per region and set up opt-out handling
- [ ] 4. Set up sending infrastructure and warm-up (before writing copy)
- [ ] 5. Research each account; write the first email and follow-ups
- [ ] 6. Pilot with a small batch, review replies, then scale slowly
- [ ] 7. Measure, prune, and iterate

1. ICP and trigger

Write the ICP as filters you can actually query:

Firmographics: B2B SaaS, 50-500 employees, US/UK, Series A-C
Role: Head of / VP Engineering or Platform lead (economic buyer or strong influencer)
Problem signal: hiring SREs, recent incident postmortem, migrating to Kubernetes
Exclusions: current customers, open opportunities, competitors, suppression list

The trigger is the reason to write now: new role (first 90 days), funding, a job post for the problem you solve, a public post, a tech change visible in their stack, an event. No trigger means weak relevance; deprioritise the account.

2. List quality

• Verify every address with a verification service before sending; drop invalid and risky results and avoid role inboxes (info@, sales@) unless the role is the target.
• Keep bounces low (practitioners commonly aim for under about 2%); high bounces signal a bought or stale list and hurt reputation.
• Store per contact: name, role, company, work email, LinkedIn URL, source, date collected, country, lawful basis, trigger, personalisation note.
• Deduplicate against CRM, current customers, and the global suppression list every time.

3. Legal basis and opt-outs

Recipient location	Default approach
United States	CAN-SPAM: honest headers and subject, identify as commercial, postal address, opt-out honoured within 10 business days
Canada	CASL needs express or implied consent; email only contacts with a recorded consent route
United Kingdom	Corporate addresses allowed with identification and opt-out; sole traders need consent; UK GDPR legitimate interests assessment for named people
EU	GDPR legitimate interests plus national ePrivacy rules; Article 14 privacy notice at first contact
Germany	Treat cold email as requiring prior consent; use other channels

For details, penalties, and record-keeping, read references/legal-by-region.md whenever the list includes recipients outside the United States or the user asks whether something is allowed. Tell the user to confirm with counsel before scaling; this is not legal advice.

Opt-out handling:

• Include a plain line such as “If this isn’t relevant, reply ‘no’ and I won’t follow up.” plus the legal footer the region needs (postal address, privacy link).
• Add a List-Unsubscribe header if the sending tool supports it.
• Treat any negative reply (“not interested”, “remove me”, angry replies) as an opt-out.
• Sync suppression across all inboxes, tools, and teammates within one business day.

4. Deliverability basics

Set up before writing copy, because a new domain needs weeks of warm-up:

• Send cold email from a separate domain (for example getacme.com or tryacme.com) that redirects to the main site, so complaints cannot damage the primary domain used for customer and transactional mail. It does not exempt you from any rule.
• Authenticate it: SPF, DKIM, and DMARC (at least p=none), with the From domain aligned. For DNS records and reputation monitoring, use the email-deliverability skill.
• Warm up each new inbox gradually over two to four weeks with low volume and real conversations before cold sends.
• Conservative caps per inbox: start around 20 to 30 cold sends a day and stay well under about 50; add inboxes rather than pushing one harder. These are practitioner norms, not provider-published limits.
• Keep Gmail spam rate below 0.3% (Google’s requirement; aim under 0.1%). Monitor with Google Postmaster Tools.
• First email: plain text, no images, no attachments, at most one link (or none), no open-tracking pixel. Tracking pixels and link redirects hurt inbox placement and open data is unreliable anyway due to privacy proxies.

5. Write the messages

First email structure (50 to 125 words):

Subject: <2-5 words, lowercase is fine, specific to them>

<Line 1: the trigger or observation about them, specific and true>
<Line 2: the problem it usually creates for people in their role>
<Line 3: one sentence of credible proof: a similar customer outcome the user can stand behind>
<Line 4: one low-friction ask>

<Name>
<Title, Company>   <opt-out line>   <postal address / privacy link where required>

Example (facts supplied by the user):

Subject: on-call after the EKS move

Hi Priya, saw the platform team posted two SRE roles right after the EKS migration
talk at KubeCon.

Teams in that spot usually find alert volume doubles before tuning catches up.
Northwind cut pages per engineer by about 40% in six weeks with our alert routing.

Worth a 15-minute look, or is on-call load not a priority this quarter?

Sam Lee, Founder, Acme
Not relevant? Reply "no" and I won't follow up.

Rules:

• Write about their situation; the product appears only as a proof line.
• Subject lines: specific, short, no clickbait, no fake “Re:”, no ALL CAPS or “Quick question” spam patterns. The subject must reflect the content (CAN-SPAM).
• Proof claims must be real and approved by the user; no invented logos or numbers.
• Read it aloud: if it sounds like a template with a name merged in, rewrite line 1.
• Personalisation tiers: tier 1 accounts get individual research; tier 2 get a segment insight (role + trigger); never fake tier 1 with merge fields like {{recent_post}}.

LinkedIn:

• Connection note: short and specific, no pitch. LinkedIn caps note length (currently around 200 characters for free accounts) and limits personalised notes; check the current limit in the UI.
• Message after acceptance: one line of context, one question, no calendar link in the first message.
• Engage with their posts genuinely before or instead of messaging.

6. Follow-up cadence

Step	Day	Channel	Content
1	0	Email	First email
2	3	Email (same thread)	New angle or useful resource, two to three lines
3	7 to 8	LinkedIn	Connection request with short note (if appropriate)
4	12 to 14	Email	Different proof point or question
5	20 to 25	Email	Close the loop: “I’ll stop here; reply if timing changes”

Stop at four or five touches total. Stop immediately on any reply, opt-out, bounce, or out-of-office with a leaving date. Never re-add a non-responder to a new sequence within a few months without a new trigger.

7. Measure

Metric	Definition	Use
Bounce rate	bounced / sent	List quality; pause above a few percent
Reply rate	replies / delivered (unique people)	Messaging and targeting together
Positive reply rate	interested replies / delivered	The primary metric
Meeting rate	meetings booked / delivered	Pipeline signal
Opt-out / complaint rate	opt-outs and spam reports / delivered	Relevance; rising = stop and fix targeting

Open rate is not a decision metric (privacy proxies and image blocking distort it). Pilot with about 50 to 100 contacts per segment before scaling, change one variable at a time (segment, trigger, first line, ask), and compare positive reply rates between batches of similar size. Do not quote benchmark reply rates as targets; use the user’s own baseline.

Gotchas

• Bought lists: high bounce, no lawful basis in many regions, and spam traps. Avoid.
• Sending from the main company domain: one bad campaign can push invoices and password resets to spam.
• Merge-field failures (“Hi {{first_name}}”) and wrong company names destroy credibility; spot-check every batch.
• Multiple teammates emailing the same account without coordination looks like spam and ignores opt-outs; use one owner per account.
• AI-written “personalisation” that misstates facts about the person is worse than none. Every personal detail must come from a source you can point to.
• Ramping volume fast after a good pilot is the most common way to burn a domain.

Verification

• Every contact has a source, country, and lawful basis recorded; suppression list applied.
• Each draft passes: true trigger, under about 125 words, one ask, opt-out line, legal footer for the region, subject matches content, no tracking pixel in email 1.
• SPF, DKIM, DMARC pass on a test send to a Gmail account (check “Show original”).
• Pilot results reviewed before scaling: bounce, positive reply, and complaint rates.

Output contract

## Outreach plan: <segment> (<date>)
ICP filters: ... | Trigger: ... | Regions and legal basis: ...
Infrastructure: sending domain, inboxes, warm-up start, daily cap per inbox
List: <n> verified contacts, source, suppression applied (yes/no)

## Sequence
| Step | Day | Channel | Subject / hook | Body |

## Personalisation notes (tier 1 accounts)
| Account | Contact | Trigger (with source link) | First line |

## Measurement
Pilot size, primary metric (positive reply rate), stop rules, review date
## Open questions for the user (proof points to confirm, regions to clear with counsel)