Stop retyping “be concise”.
Stylespec puts your response style one click from every AI chat box — real style guides, not moods — and sends the same styled prompt to a second assistant when one answer is not enough.
No broad site access. No network requests. No accounts. Open source.
The Chrome Web Store listing is in review. Until it is live, the folder loads unpacked in Chrome, Brave, Edge or Arc with no build step.
The problem
Every assistant has the same voice
Long. Hedged. Bulleted. It repeats your question back to you, then offers to help further.
That is the right default when one voice has to serve everybody. It is the wrong one for you, on the fortieth prompt of the day, when you already know what you asked.
So you type “be concise” at the end. And “no bullet points”. And “just the code”. Every time, in every tool, and never once does it stick.
Great question! Centering a div is one of those
tasks that seems simple but has a surprising amount of nuance
depending on your specific use case and the CSS layout model you
are working within.
Before we dive in, it is worth understanding that there are several approaches available to us, each with their own tradeoffs in terms of browser support, maintainability, and flexibility. Let us walk through them together.
First, we should consider whether you need to center horizontally, vertically, or both. This distinction matters because historically, horizontal centering has been straightforward while vertical centering was notoriously difficult…
Option 1: Flexbox. Flexbox is a one-dimensional layout method that has become widely supported across all modern browsers…
display: grid; place-items: center; on the parent.
Parent needs a height for vertical centering to be visible. Use
place-content: center instead if the child should
keep its own size.
How it works
Three seconds, once
Type your question
Anywhere. The pill appears next to the box on your first keystroke.
Pick a style
23 built in, or write your own. Stack length, formality, audience, emoji and language on top.
Send
The instruction lands in your draft, in plain sight, where you can still edit it.
Nothing is added behind your back. You see every word before it goes.
Styles
The style guides you actually wanted
Five of them are published standards rather than preferences. Those apply the written rules of each guide.
Simplified Technical English
ASD-STE100One approved meaning per word, sentences under twenty words, one instruction per sentence.
BLUF
StandardBottom line up front, at most three supporting bullets, under 150 words.
Plain language
plainlanguage.govShort sentences, common everyday words, active voice, most important information first.
No LLM tells
No em dashes. No “delve”, no “tapestry”, no “testament to”. No closing paragraph summarising the paragraph above it.
Put the fries in the bag
Answer first. No preamble, no restatement, no “great question”. When it has answered, it stops.
Terse senior engineer
Code and commands first. Prose only where it carries something the code cannot.
Explain it like I’m 5
Everyday words and one concrete analogy, held all the way through.
Devil’s advocate
Assumes your idea is flawed and finds the strongest reasons why. Leads with the failure most likely to actually happen.
Socratic tutor
Refuses to answer. Asks the one question that moves you forward, and waits.
Plus Feynman, Hemingway, the Google and Microsoft documentation house styles, Academic, Legalese, Financial analyst, Checklist, Bullets only, Table first — and Pirate, Shakespearean, Corporate LinkedIn and Gen Z, which stay accurate underneath the accent.
Write your own in thirty seconds. Name it, type the instruction, star it to pin it to the top.
Everywhere
Your styles outlive your favourite assistant
Those are the ones recognised by name. Stylespec finds chat boxes by how they look and behave, not by a list of domains — so your company’s internal assistant, your self-hosted front-end, and whatever launches next month generally work on day one.
Multiplexer
One tick, and a second assistant is already answering
Some questions deserve more than one answer, and getting a second one means selecting your prompt, copying it, opening a tab, pasting it, sending. Every time.
Tick ChatGPT or Claude in the picker and just send. Each opens in a background tab to the right, already asking your question, with your style instruction attached. You never leave the tab you were in.
It starts where you already are.
The side-by-side tools are a destination — you type into their box instead of your assistant’s, abandoning the conversation you were already in. Stylespec works the other way round. It sends the draft you had already written, from the site you were already on.
The comparison is actually fair.
Two answers are only comparable if the prompt was identical, and prompt shape drives answer shape more than people expect. The same style instruction goes to both, so you are comparing the models — not one model’s default verbosity against another’s.
ChatGPT and Claude built in; add any assistant that takes a prompt in its address, including an internal one. Answers arrive in normal tabs on each assistant’s own site — there is no split-screen pane, and nothing is read back or scored for you.
Against the alternative
“Doesn’t ChatGPT already do this?”
It does, inside ChatGPT, for every message, until you go and change it. Which is genuinely useful, and not the same thing.
| Vendor settings | Stylespec | |
|---|---|---|
| Works on | One assistant | Every assistant |
| Applies | To every message | To this message |
| Switching style | Open settings, edit, save | One click |
| Keeping them in sync | Three tools, three copies | One library |
| What it adds | Hidden in the system prompt | Visible in your draft |
Most people reading this have three assistants open right now. That is the whole argument.
Privacy
It cannot run where you have not let it
Install almost any extension like this and Chrome warns you it can “read and change all your data on all websites”. Stylespec never asks for that.
It runs on the assistant sites you would expect, and nowhere else. Anything beyond that is one explicit click, granting one site at a time.
That is unusual for anything that talks to several assistants at once, which normally needs standing access to all of them. Stylespec passes the prompt in the address bar instead, so for most targets it injects nothing at all.
No network requests. No analytics, no telemetry, no accounts, no remote code. Your prompts are read by the site you typed them into and by nothing else. Your styles sit in your browser’s own sync storage.
Your AI chats are among the most revealing text you produce — half-formed ideas, code you have not shipped, things you would not ask a colleague. An extension that watches that is a serious thing to install. This one is incapable of it, and you do not have to take that on faith.
FAQ
Questions people actually ask
Does this make the AI smarter?
No, and anything claiming otherwise is selling something. The substance stays the model’s. Stylespec changes the shape it arrives in — which, most days, is the part costing you time.
How is the Multiplexer different from the side-by-side comparison extensions?
Those are dashboards, and several are good. If you want every model in one window with the answers collected for you, use one of them. Stylespec is the assistant you already had open, plus a tick box that saves you the copy and paste — and it is the only one that sends the same style instruction to both sides, which is what makes two answers comparable at all.
Is this an ASD-STE100 compliance tool?
No. ASD-STE100 has two parts, a set of writing rules and a controlled dictionary. Stylespec applies the writing rules as an instruction to the model. It does not check output against the official dictionary, so it is a drafting aid rather than a compliance checker.
Can it send to Gemini, Perplexity or Grok?
Add them in Settings, along with anything else that takes a prompt in its address. The built-in list is short deliberately: those addresses are undocumented and can change without warning, so each one shipped is a maintenance promise rather than a line on a feature list.
Will it break when ChatGPT redesigns?
Probably not. Site-specific rules are only confidence boosts; the detection underneath looks at the shape of the page. That is why unreleased assistants tend to work already.
It didn’t appear on my assistant.
Open the toolbar popup and hit Turn it on here. Chrome will ask permission for that one site, and detection takes over from there. Then please open an issue — a miss is a bug worth fixing, and fixing one usually fixes a class.
Doesn’t a longer prompt cost more tokens?
A style block is 40 to 70 words and removes several hundred words of output. It nets out cheaper, in tokens and in reading time.
Can I use it for work?
Yes. No data leaves the machine, so there is nothing for a security review to object to. Point them at the source.
Firefox? Safari?
Not yet. The code is standard Manifest V3, so a Firefox build is mostly packaging. Open an issue if you want it and it moves up the list.
What does it cost?
Nothing. No paid tier is planned, no accounts, no upsell.
Set the style once. Use it everywhere.
Open source · No broad site access · No tracking