Skip to content

Checks

Voice check

Checks any draft against your company's own voice rules and 3 copy tests, then returns pass or fail with a fix for every line that doesn't sound like you.

Other tools

Add this skill to Claude

Claude takes skills as a ZIP file. Your download has started; if it didn't, use the button below.

  1. 1In Claude, open Customize, then Skills.
  2. 2Choose +, then Upload a skill, and pick voice-check.zip.
  3. 3Ask for it by name, or let Claude use it when a task matches.

Skills need code execution turned on in Claude's settings. On a Team or Enterprise plan, an admin can share it with everyone.

Download the folder

Try it now, without installing

See the skill work on a made-up input before you add it. Nothing is installed, and you can swap in your own input.

Open in the Claude app

Opens a new chat in the Claude desktop app with the prompt filled in. Nothing is sent until you press enter.

Includes the whole SKILL.md and the sample input. Paste it into Claude, ChatGPT or any chat.

Influenced by
3 copy tests
Needs
The draft
Works with
Claude, ChatGPT, Codex, Cursor and any agent that reads SKILL.md
Version
1.0.0 · Last checked · License MIT
Maintained by
Sabih AhmedContent
SkillStart here

Voice check

Example output

Voice check on a homepage draft
LineIssueCall
4Word you never useFix
934-word sentenceSplit
12Sounds like youKeep

Checks any draft against your company's own voice rules and 3 copy tests, then returns pass or fail with a fix for every line that doesn't sound like you.

The skill

The whole skill, in plain text.

Nothing is hidden. Read it, change it to fit your team, or take the whole folder.

Scripts
Optional Python scripts for the counting. It works without them and says which way it ran.
Network
Doesn't need the internet.

3 test cases

voice-check/SKILL.md

In the folder

  • voice-check/SKILL.md
  • voice-check/references/ai-patterns.md
  • voice-check/scripts/voice_lint.py
---name: voice-checkdescription: >-  Checks any draft (page, post, ad, email) against the company's own voice rules from company-profile.md, a list of  patterns that make copy read as machine-written, and 3 copy tests, then returns PASS or FAIL with line-level fixes.  Use it before anything goes out, or as the check step inside any skill that writes copy. Not for checking  whether claims are true (use claim-check) or for reviewing a whole landing page's structure (use  landing-page-critic).license: MITcompatibility: >-  Works in Claude, ChatGPT, Codex, Cursor and any agent that reads SKILL.md. Python 3 is optional, for the  mechanical pass.metadata:  author: SHAKEUP  version: "1.0.0"  stage: check  plugin: start-here  homepage: https://www.shakeuphq.com/skills/voice-check--- # Voice check Finds every line that doesn't sound like the company, says why, and offers the fix. ## Purpose - One job: check one draft against the company's voice and return fixes.- Stage: check. It ends in PASS or FAIL with severities.- It suggests fixes; it doesn't rewrite the whole draft. A rewrite belongs to the skill that made the draft. ## Use when / not | Situation | Use || --- | --- || A draft is ready and must sound like the company | this skill || Another skill reaches its check step on copy | this skill || You need to know if a number or claim is backed | `claim-check` || You want a structural review of a landing page | `landing-page-critic` || You want to test if buyers understand the message | `message-test` | ## Inputs | Input | Required | Ways to provide, in order | If missing || --- | --- | --- | --- || The draft | yes | a file the agent can read; pasted text | ask for it || What it is and who reads it | no | the reader's note; infer from the draft | infer, and say what you assumed || Voice rules | no | `## Voice` in `company-profile.md`; pasted rules | run on the AI patterns and copy tests only; coverage `partial` | ## Profile read `## Voice` (point of view, tone, reading level, words to use, words to avoid, lines that sound right), `## Buyers` (who reads it), `## Claims and proof` (only to avoid flagging an approved claim as vague), and `## Learnings` entries for `voice-check`. No profile: ask for 3 to 5 words that describe the voice and any words the company never uses, in one message. If the reader has none, continue with coverage `partial`. ## Steps 1. **Apply learnings** for this skill (accepted rules override the defaults below). 2. **Mechanical pass.** Run `scripts/voice_lint.py`:    ```   python3 scripts/voice_lint.py --draft draft.md --profile company-profile.md   ```    Without Python, check by hand, line by line: em dashes and en dashes; every phrase under "Words to avoid"; every phrase in `references/ai-patterns.md`; sentences over 30 words; paragraphs over 4 sentences; numbers from two up written as words; consecutive sentences with the same first word. Say which way you ran it. 3. **Voice pass.** Read the draft as its reader would. For each paragraph, check:    | Check | Flag when |   | --- | --- |   | Point of view | It switches between "we", "I" and "you" against the profile |   | Tone | A line breaks one of the profile's tone words (a "direct" voice that hedges; a "warm" voice that lectures) |   | Reading level | Jargon the profile's reader wouldn't use, with no explanation |   | Sounds like the company | It reads unlike the profile's "Lines that sound right" | 4. **Copy tests.** Apply 3 tests to every claim and headline:    | Test | Fails when |   | --- | --- |   | Can I picture it? | Abstract words with no image, number or example ("drive growth", "better outcomes") |   | Can it be proved wrong? | Nothing in it could be checked ("the best platform") |   | Could no one else say it? | A competitor could put it on their site unchanged | 5. **Constructions.** Check the list under "Constructions to watch" in `references/ai-patterns.md`. 6. **Write the fix.** For each finding: the line, the rule, the problem in a few words, and a rewrite that fixes only that problem in the company's voice. Never fix by adding a claim the profile doesn't support. ## Check and verdict Severity: `critical` (a dash; a word to avoid; a point-of-view switch; a line that contradicts the profile), `warning` (an AI pattern; a long sentence or paragraph; a failed copy test on a headline), `suggestion` (number words; repeated openers; a failed copy test in body copy). | Verdict | Condition (first match wins) || --- | --- || `BLOCKED` | No draft || `FAIL` | 1 or more critical findings, or 5 or more warnings || `PASS WITH NOTES` | Warnings or suggestions only, fewer than 5 warnings || `PASS` | No findings | No named stops. The calling skill decides what to do with a FAIL. ## Learn When the reader rejects a finding as wrong for their voice ("we do say 'unlock' in this market"), write a `proposal` learning to remove that rule for this company. When the reader adds a word to avoid, add it to the profile's `### Words to avoid` and record a `fact` learning. ## Output ```Verdict: <PASS | PASS WITH NOTES | FAIL | BLOCKED> · Coverage: <complete | partial (no voice rules in profile)> · Mechanical pass: <script | by hand>Counts: <n> critical, <n> warnings, <n> suggestions | Line | Severity | Rule | Problem | Fix |(critical first) Strongest line: "<the line that most sounds like the company>"``` Then the handoff block. ## Hand off ```handoffskill: voice-checkversion: 1.0.0verdict: FAILcoverage: completefor: <draft name>outputs:  critical: 2  warnings: 3  fixes_ready: truenext: [claim-check]open: []``` The skill that made the draft reads `critical` and the fix table, applies the fixes and runs this check again (its own loop, at most 2 rounds). ## Failure modes | Problem | Do this || --- | --- || The draft is in another language than the profile | Check dashes, numbers and structure only; say voice rules weren't applied || The profile's words to avoid include a word the product name contains | Don't flag the product name || A quote from a customer breaks the voice | Quotes are theirs: don't flag inside quotation marks || The draft is a table or ad copy with fragments | Fragments are fine in labels, buttons and ad headlines; don't flag them there | ## Influenced by - Harry Dry's three copy tests (Marketing Examples): https://www.demandcurve.com/newsletters/growth-newsletter-202- SHAKEUP's copy lint, which fails a build on dashes, banned words, sentences over 30 words and paragraphs over 4 sentences.

How it thinks

Inputs in, a verdict out, and a stop only where you decide.

What we built

Our own site runs a copy lint that fails the build on dashes, banned words, sentences over 30 words and paragraphs over 4 sentences. This skill brings that discipline to any team, with your rules instead of ours.

It reads your company profile if you have one, or asks for only what it needs. Then it works through its steps, checks its own work and ends in a verdict.

The steps

  1. 01

    The mechanical pass

    Its script flags dashes, your words to avoid, phrases that read as machine-written, long sentences and paragraphs, and spelled-out figures. Without Python, it checks the same list line by line.

  2. 02

    The voice pass

    It reads the draft as your buyer would and checks point of view, tone, reading level and whether each paragraph sounds like your best lines.

  3. 03

    3 copy tests

    Every headline and claim has to pass all 3. Can a reader picture it, could it be proved wrong, and is it something only you could say?

  4. 04

    A fix for each finding

    Each finding gets a rewrite that fixes only that problem, in your voice, with no new claims added.

What you need

  • Required

    The draft

    A file your agent can read → Text you paste

  • Optional

    What it is and who reads it

    Your note → Inferred from the draft

  • Optional

    Your voice rules

    The Voice section of your company profile → Rules you paste

What it can return

  • PASS

    Nothing to fix.

  • PASS WITH NOTES

    Warnings or suggestions only, fewer than 5 warnings.

  • FAIL

    Any critical finding, such as a word you never use or a switch in point of view, or 5 warnings or more.

  • BLOCKED

    There's no draft.

Where it stops for you

It runs through without stopping. You review the result.

Example output

What you get back.

A run for a made-up company. Your output depends on your inputs.

Here's a run on a homepage draft for a made-up B2B SaaS company whose voice is direct, practical and calm.

Verdict

FAIL

Coverage: complete

Mechanical pass: script

LineSeverityRuleProblemFix
1CriticalWord to avoid"revolutionary""Our attribution platform..."
1CriticalWord to avoid"single source of truth""...figures your RevOps team already trusts"
1CriticalDashA dash splits the sentenceEnd the sentence and start a new one
3WarningLong sentence41 wordsSplit after "campaign"
3WarningCopy test"every touchpoint" can't be picturedName the 3 channels it reads
Strongest line

"Live in 2 weeks, with your stages left as they are."

The handoff block the next skill reads
skill: voice-check
version: 1.0.0
verdict: FAIL
coverage: complete
for: homepage draft
outputs:
  critical: 3
  warnings: 2
  fixes_ready: true
next: [claim-check]
open: []

What good looks like

  • Every finding has a line, a rule and a fix
  • Fixes stay in your voice and add nothing new
  • Quotes from customers are left alone
  • Your strongest line is named, so you know what to keep

Install

Use it in the tool you already have.

Pick your tool. Each tab has the shortest path we know of.

  1. 01
    Download voice-check.zip.
  2. 02
    In Claude, open Customize, then Skills.
  3. 03
    Choose +, then Upload a skill, and pick the ZIP. Code execution must be on in your settings.
  4. 04
    Ask for it by name, or let Claude use it when a task matches.

Questions

Before you add it.

Do I need written voice rules?

No, but it works best with them. Without rules, it checks the general patterns and the copy tests, and tells you the result is partial.

Will it rewrite my draft?

No. It gives a fix for each line it flags, and the skill or person that wrote the draft applies them.

Does it flag fragments in ad copy?

No. Fragments are fine in headlines, labels and buttons. It only flags them in body copy.