First kit draft
hey Sam, the poster draft is ready. can you check the headline by Tuesday? thanks, Casey
Voice Mining Kit / Writing with AI
Make it sound
like you.
Turn your own writing into a voice skill for Claude Code or Codex. You can read every rule and change anything that feels off.
$129once
For local Claude Code or Codex. Python 3.10+ and your own AI account required.
See what the kit actually makes
BZK / SALES DESK
FurbyUnsupervised sales department
Ask the little guy.
Furby / Michael's demo voice
Ask about the kit
AI chat. Messages go to Anthropic. Keep personal details out.
Furby uses a version of my voice instructions. The kit walks you through building yours.
Okay, show me.
I ran the kit on two fictional writers. Here's what changed when it checked the first draft.
After AI review
hey Sam, the poster draft is ready. can you check the headline by Tuesday? the print colours are still under review. thanks, Casey
The review restored the missing colour status.
Fictional writing. Actual model outputs. Human voice review is still pending.
Download the full runSee the source and review notes
The brief
Write an email to a client named Sam. The poster draft is ready. Ask Sam to check the headline by Tuesday. The print colours are still under review. Do not promise a print date or invent a link.
What the review found
The kit review restores the required fact that print colours are still under review. Its ready status and Tuesday request remain.
The baseline review leaves its unsupported promise to be in touch unchanged. That comparison draft still needs correction, despite the model saying no change was warranted.
A sample of their writing
hey [client], the draft's ready. give the examples a look when you've got a minute. thanks, Casey
One of 6 learning samples. 2 more were kept aside for review.
From the generated skill
- **Greeting:** "hey [client]," in every sample (C1, C2, C3, C5, C6, C7). Lowercase throughout. - **Signoff:** "thanks,\nCasey" in every sample. "Casey" is capitalised; the word "thanks" is lowercase. - **Body length:** One short paragraph per message in all six samples. Typically one to two sentences.
These rules still need corrections.
The generated skill and drafts preserve hey, a casual body, and thanks with a capitalised signature.
Its capitalization rule contradicts the lowercase sample bodies. The evidence also wrongly says contractions appear in every sample and cites you'd in C5. These generated observations need correction before reuse.
The reserved C4 and C8 share the lowercase body and short conversational structure. This is an agent observation on synthetic writing, not an author verdict.
With an ordinary prompt
hey Sam, the poster draft is ready. can you check the headline by Tuesday? the print colours are still under review, so i'll be in touch on that separately. thanks, Casey
Both versions used claude-sonnet-4-6, the same writing samples, and the same brief. Each received an AI review. Original outputs are preserved in the full download.
This comparison still needs a correction.
The reviewed baseline still adds 'i'll be in touch on that separately', an unsupported follow-up promise. Remove it before using this draft.
From this to this.
Before
I spent 20 minutes troubleshooting the microphone before realizing I had muted it. It's working now.
After
Bruh I spent 20 minutes trying to fix my mic. It was muted the whole time 😭
Working now.
The whole workflow,
in files.
Open the start guide and follow one worked example. Then bring your own writing. You get three agent skills, Python helpers, and a way to compare the drafts.
- 01
Bring your writing
Start with emails or posts you've written. Keep a few aside for comparison.
- 02
Build your skill
Turn those examples into a voice skill you can read and change.
- 03
Check the drafts
Compare the results and catch the phrases you keep having to fix.
What you keep when you're done
Your finished folder includes the voice skill, the examples behind its rules, and a checker configuration you control. Install it in your agent and keep adjusting it as you use it.
Everything is editable. You can use what you build on your own projects, including commercial work.
One payment. Local Claude Code or Codex, Python 3.10+, and your own AI account required.
Made by Michael Dishmon
Yeah, I use this stuff.
I work across design and development. This kit comes from the way I build and use my own voice skill. The little things matter, including knowing when a sentence needs a bruh.
See what I buildFeedback on my design and consulting work
“Michael is a kickass dude. I really enjoyed our conversation, we went through an awesome persona building process that really helped me hone in on some of my marketing efforts.”
“I greatly enjoyed working with Michael. He puts a lot of thought and heart into his work and has truly great ideas. I found his designs to be high caliber and intuitive. He is professional, creative, and technical. A great combination, plus an all-around fantastic person to work with.”
A few details.
What do I need to run it?
Local Claude Code or Codex with file access, Python 3.10+, and your own writing. Your agent account and model usage are separate. The guide uses macOS or Linux commands; Windows users can use WSL or extract the ZIP themselves. Both Python helpers are for English.
What comes in the download?
Three agent skills, two writing helpers, a setup check, and the start guide. The worked example includes fictional source writing, generated skills, and the original comparison drafts. You run the kit yourself; personal setup and writing services are outside the purchase.
Do you get my writing?
The kit doesn't send your writing to me. Pasting messages works, so connecting email is optional. Your agent processes text through its model provider. Check its privacy settings, remove confidential details, and keep your samples out of shared repositories.
How do I know it sounds like me?
Set some writing aside before building the skill. Use the same brief to compare drafts with and without it, then read both alongside your originals. You decide what sounds right. The checker catches only the rules you turn on.
Can I change the rules?
Yep. The files are editable, and the checker starts with every optional check turned off. Your writing supplies the examples. Read through what the agent builds, then change anything that feels off before installing it.
I already give my AI writing samples. Why buy this?
If that gets you drafts you're happy with, keep doing it. This kit gives you a repeatable process for choosing samples, recording the evidence behind each rule, and testing the skill on writing you kept aside. You get an editable skill and a checker you control.
How much writing should I start with?
A useful starting point is 12 to 20 complete messages in one kind of writing. Keep roughly a quarter aside for comparison. Setup time depends on your source material and tools, and you should expect to read and adjust what the agent produces.
What are the license and refund terms?
Use the kit on your own projects, including commercial work. Reselling or redistributing the kit is outside the license. Digital sales are final once access is granted. If a file is defective or doesn't match the description, contact me for a fix, replacement, or refund under the site's terms.
Give it your own words.
Start with writing you already have. Keep the finished skill and use it on your own projects, including commercial work.
$129one payment
Download after payment, with a link by email too. Your AI account and model usage are separate.