New: Explore the AI Stack for Claude skills, outbound automations, and operator workflows
Claude Skills8 min read·

Cold-Email Copy Skill: Spam Guard

A copy-paste Claude or Codex skill that blocks spam-trigger wording, strips fake urgency from cold emails, and forces plain-language rewrites before copy goes out.

Cold-Email Copy Skill: Spam Guard

Most cold email copy gets judged twice: once by the prospect and once by the inbox. A line can sound strong to a human and still look risky to filters, QA reviewers, or internal deliverability checks.

That is where a dedicated Spam Guard skill becomes useful. Instead of hoping you remember every risky token, phrase, and formatting habit, you give Claude or Codex a standing ruleset that screens subject lines, openers, follow-ups, and CTAs before the copy ever leaves your workflow.

This article gives you the exact skill structure to paste into Claude or Codex, along with the logic behind it: what to ban, how to rewrite risky wording, how to handle edge cases, and how to force the model toward lower-hype language by default.

What This Skill Is For

This is an always-on spam and deliverability layer for outbound copy. You paste it into Claude or Codex as a reusable skill so every generated output gets screened before you approve it.

The point is not to make copy dull. The point is to stop the model from drifting into language that reads promotional, scammy, pressure-heavy, or unnecessarily polished in an AI way.

When in doubt, the skill should prefer the lower-hype rewrite instead of trying to defend borderline wording.

When These Rules Apply

  • Apply them to subject lines, openers, second lines, follow-ups, and CTA-style copy.
  • If a deliverability checker or QA review flags a word or phrase, treat it as banned going forward unless the user explicitly approves an exception.
  • If a line sounds like an ad, coupon, scam, phishing email, or fake-urgent push, rewrite it.
  • If a bump is filler-heavy or vague, replace it with a direct next-step question or a clear closeout line.
  • If a reply acknowledgement feels low-status or unnecessary, remove it and move straight to the next useful question.

What This Skill Blocks

The guardrails cover four broad types of risk:

  1. 1Internal QA banned single words such as `access`, `deal`, `urgent`, `invoice`, `purchase`, and similar tokens that repeatedly trip review.
  2. 2Internal QA banned short phrases such as `following up here`, `great fit`, `circle back`, and other overused outbound filler.
  3. 3Broader promotional and pressure-heavy wording such as `100% free`, `act now`, `limited time`, `buy now`, `guaranteed results`, and similar spam-adjacent language.
  4. 4Phishing-style and security-warning language such as `password reset`, `verify identity`, `account update`, and other terms that make a cold email look like a spoofed system message.

It also bans irrelevant blacklisted categories that should never show up in the user's copy at all, including gambling, miracle-cure, adult, or pseudo-medical language.

How The Rewrite Logic Should Work

  • Rewrite hype into plain, observational language.
  • Replace pressure with permission.
  • Replace promotional wording with specific business language.
  • Rewrite fuzzy terms like `tight`, `fit`, `access`, or `problem` so the operational meaning is explicit.
  • Simplify any sentence that sounds polished in an AI way until it reads like a person speaking plainly.
A good rewrite should sound calmer, more specific, and less performative than the original.

Nuance Rules That Matter

Root tokens are still unsafe

  • Punctuation does not make a banned token safe.
  • Hyphenating a banned token does not make it safe.
  • Splitting or joining the word does not make it safe if the root token is still obvious.
  • Treat close variants as banned too when the risky root token is still clearly present.

Company names need deterministic rewrites

  • If a banned word appears inside a company name, do not remove the company reference entirely if the line still needs the name.
  • Rewrite the displayed company name so the banned token is gone from standalone form.
  • First choice: remove the banned token if the remaining name is still recognizable.
  • Only abbreviate or compress when removing the token would make the name unclear.
OriginalRewrite
Access Brand CommunicationsAB Communications
Calcon Mutual MortgageCalcon Mutual
Buckeye InsuranceBuckeye
Coming Soon New YorkComing NY

Formatting And Style Bans

  • No em dashes.
  • No ALL CAPS.
  • No multiple exclamation marks.
  • No greeting prefix before the first name such as Hi, Hello, or Hey.
  • No third-person company references such as `[Company] offers` or `[Company] helps`.
  • No fake urgency, misleading subject lines, excessive links, or promotional formatting.

Safe Replacement Patterns

AvoidUse Instead
`free consultation``open to a short conversation`
`special offer``what we’re seeing in the market`
`act now``if relevant, happy to send details`
`guaranteed results``this may be relevant depending on your situation`
`click here``let me know and I can send it over`
`limited time``not sure if this is timely for you`
`increase revenue``help support liquidity` when that is actually true

Copy-And-Paste Skill

Paste this directly into Codex or Claude as the skill definition:

Cold-Email Copy Skill: Spam Guard
Always-on spam and deliverability guardrails for every generated output.

Use these rules to decide what to avoid, how to rewrite risky wording, and when to treat a word or phrase as unsafe.

When these rules apply:
- Apply them to subject lines, openers, second lines, follow-ups, and any CTA-style copy.
- If a deliverability checker or QA review flags a word or phrase, treat it as banned going forward unless the user explicitly approves an exception.
- When in doubt, choose the lower-hype rewrite instead of trying to defend borderline wording.

What to avoid - internal QA banned single words:
- `get`, `bank`, `credit`, `access`, `open`, `compare`, `problem`, `now`, `billing`, `deal`, `finance`, `financial`, `claims`, `insurance`, `mortgage`, `soon`, `new`, `performance`, `freedom`, `home`, `sales`, `medical`, `urgent`, `life`, `marketing`, `investment`, `diagnostics`, `friend`, `cash`, `invoice`, `extra`, `purchase`

What to avoid - internal QA banned short phrases:
- `off chance`, `one time`, `all good`, `following up here`, `last note from me here`, `great fit`, `bumping this once`, `just following up once`, `circle back`, `one more quick follow-up`, `keep this open`, `compare notes`, `compare notes live`, `appreciate the reply`

What to avoid - broader high-risk promotional or pressure wording:
- `$$$`, `50% off`, `100% guaranteed`, `100% free`, `100% off`, `100% satisfied`, `access now`, `act fast`, `act immediately`, `act now`, `action required`, `affordable deal`, `amazing`, `amazing deal`, `amazing offer`, `apply here`, `apply now`, `avoid bankruptcy`, `bargain`, `best bargain`, `best deal`, `best offer`, `best price`, `best rates`, `big profit`, `bonus`, `buy now`, `buy today`, `call now`, `can't live without`, `cash bonus`, `cash out`, `claim now`, `claim your discount`, `click`, `click below`, `click here`, `click this link`, `contact us immediately`, `deal ending soon`, `discount`, `don't delete`, `double your money`, `double your wealth`, `drastically reduced`, `earn`, `earn cash`, `earn extra income`, `earn money`, `easy income`, `exclusive deal`, `expires today`, `extra cash`, `extra income`, `fantastic`, `fantastic offer`, `fast cash`, `final call`, `for free`, `free access`, `free consultation`, `free gift`, `free membership`, `free money`, `free quote`, `free trial`, `full refund`, `get it now`, `get out of debt`, `get started now`, `giveaway`, `great news`, `guaranteed deposit`, `guaranteed results`, `hurry up`, `important information`, `immediately`, `increase revenue`, `increase sales`, `incredible deal`, `instant earnings`, `instant income`, `instant savings`, `investment advice`, `join millions`, `limited time`, `lowest price`, `make money`, `million dollars`, `money-back guarantee`, `must read`, `no catch`, `no cost`, `no obligation`, `no strings attached`, `once in a lifetime`, `only $`, `only available here`, `order now`, `order today`, `please read`, `price protection`, `profits`, `promise`, `pure profit`, `quote`, `risk-free`, `satisfaction guaranteed`, `save $`, `save big money`, `save up to`, `sign up free`, `special invitation`, `special offer`, `special promotion`, `supplies are limited`, `take action now`, `the best`, `this won't last`, `thousands`, `time limited`, `today`, `top urgent`, `trial`, `unbeatable offer`, `unbelievable`, `unlimited`, `urgent`, `what are you waiting for?`, `while supplies last`, `why pay more?`, `will not believe`, `winner announced`, `wonderful`, `you are a winner`, `you will not believe your eyes`

What to avoid - phishing-style or security-warning language:
- `access your account`, `account update`, `activate now`, `change password`, `click to verify`, `confirm your details`, `confidential information`, `data breach`, `download now`, `final notice`, `important update`, `immediate action required`, `install now`, `last warning`, `log in now`, `new login detected`, `password reset`, `payment details needed`, `phishing alert`, `security breach`, `security update`, `update account`, `verify identity`, `warning message`

What to avoid - irrelevant blacklisted categories that should never appear in the users copy:
- `100% natural`, `adult content`, `bet now`, `blackjack`, `casino bonus`, `cure for`, `diet pill`, `doctor recommended`, `fat burner`, `fast weight loss`, `free chips`, `free spins`, `gamble online`, `guaranteed weight loss`, `jackpot`, `lottery winner`, `medical breakthrough`, `miracle cure`, `natural remedy`, `no prescription needed`, `online betting`, `online casino`, `online pharmacy`, `pain relief`, `poker tournament`, `prescription drugs`, `reverse aging`, `risk-free bet`, `safe and effective`, `scientifically proven`, `secret formula`, `slots jackpot`, `spin to win`, `vip offer`, `weight loss`, `winning numbers`, `xxx`

How to avoid them:
- Rewrite hype into plain, observational language.
- Replace pressure with permission.
- Replace promotional wording with specific business language.
- If a line sounds like an ad, coupon, scam, phishing message, or fake urgency, rewrite it.
- If a bump sounds filler-heavy or vague, replace it with a direct next-step question or a clear closeout line.
- If a value line uses fuzzy wording like `tight`, `fit`, `access`, or `problem`, rewrite it so the operational meaning is explicit.
- If a reply acknowledgement sounds low-status or unnecessary, remove it and go straight to the next useful question.
- If a sentence sounds polished in an AI way, simplify it until it reads like a person speaking plainly.

Nuance rules for when a token is still unsafe:
- Punctuation does not make a banned token safe.
- Hyphenating a banned token does not make it safe.
- Splitting or joining the word does not make it safe if the root token is still obvious.
- Treat close variants as banned too when the risky root token is still clearly present.
- Examples: `cash-cycle` is still `cash`; `invoice-line` is still `invoice`; `extra-room` is still `extra`; `purchase-cycle` is still `purchase`.

Nuance rules for company names:
- If a banned word appears inside a company name used in copy, do not drop the company reference entirely if the line still needs the name.
- Rewrite the displayed company name so the banned token is gone from standalone form.
- Prefer the simplest deterministic rewrite that keeps the name recognizable.
- First choice: remove the banned token if the remaining name still reads clearly.
- Only abbreviate or compress when removing the token would make the name unclear.
- Examples: `Access Brand Communications` -> `AB Communications`; `Calcon Mutual Mortgage` -> `Calcon Mutual`; `Buckeye Insurance` -> `Buckeye`; `Coming Soon New York` -> `Coming NY`.

Formatting and style bans:
- No em dashes.
- No ALL CAPS.
- No multiple exclamation marks.
- No greeting prefix before the first name such as Hi, Hello, or Hey.
- No third-person company references such as '[Company] offers' or '[Company] helps'.
- No fake urgency, misleading subject lines, excessive links, or promotional formatting.

Safe replacement patterns:
- Instead of `free consultation`, use `open to a short conversation`.
- Instead of `special offer`, use `what we’re seeing in the market`.
- Instead of `act now`, use `if relevant, happy to send details`.
- Instead of `guaranteed results`, use `this may be relevant depending on your situation`.
- Instead of `click here`, use `let me know and I can send it over`.
- Instead of `limited time`, use `not sure if this is timely for you`.
- Instead of `increase revenue`, use a precise business outcome such as `help support liquidity` when that is actually true.

Final check before approving copy:
- Scan the subject line for spam-trigger wording.
- Scan the body for banned or high-risk wording.
- Rewrite any hype-heavy or pressure-heavy line into plain language.
- Remove fake urgency.
- Keep the email sounding like a credible real person, not a promotion.

Final Check Before You Approve Copy

  • Scan the subject line for spam-trigger wording.
  • Scan the body for banned or high-risk wording.
  • Rewrite hype-heavy or pressure-heavy lines into plain language.
  • Remove fake urgency.
  • Keep the email sounding like a credible real person, not a promotion.

Final Thought

A lot of deliverability problems start long before technical setup breaks. They start in the copy layer, where the message quietly picks up ad language, pressure language, or filler that no serious operator would say out loud.

This skill gives Claude or Codex a harder edge on that problem. It does not just tell the model what to avoid. It tells it how to think: lower hype, more clarity, fewer risky tokens, and cleaner rewrites when the first draft drifts.

Spam Guard skill reference for cold email copy

Need cold email volume?

Done-for-you mailboxes for outbound

InfraSuite is built for teams that rely on cold email as a core revenue channel and need stable, high-performing Outlook mailboxes. You subscribe to a proven Microsoft-based sending environment that's already configured for cold outreach — provisioning, DNS, mailbox setup, and deliverability hygiene handled for you.

  • Stable inbox placement across Outlook and Google
  • Fewer resets, fewer domain swaps
  • Capacity ready when clients sign
  • Calm, competent support when something looks off
Learn more at InfraSuite

Frequently asked questions