Should I vibe code
Flag readability problems and offer optional AI rewrites inside a distraction-free editor
The highlighting is a word list and a syllable counter. What you're subscribing to is somebody else's rewrite button.
?
Their verdict, the Editor Plus price and the build-time estimate come from their entry, MIT-licensed. Checked 2026-08-04.
?
Our verdict, the regret score and everything below it. Editorial and unsponsored — nobody can pay to be moved.
The honest answer
why the verdict is what it is
The rules behind the coloured highlights are public and decades old: a grade-level formula, an adverb count, a handful of passive-voice patterns, and a list of long words with shorter alternatives. Any agent reproduces that in an afternoon, and because you are writing the rules yourself you can bend them towards your own style guide — which is the actual reason to build it, rather than saving ten dollars. Two honest caveats, neither dangerous. Your grade-level number will not match theirs, because syllable counting is a heuristic and every implementation disagrees at the edges, so treat it as a relative signal rather than a score. And the Plus half is an LLM call, which means the draft you were editing privately on your own machine now leaves it. That trade is fine when you choose it and irritating when a helpful autosave makes the choice for you.
What actually breaks
not "if". the specific failures.
- Your grade-level number quietly disagreeing with every other tool's, because syllable counting is guesswork and each implementation guesses differently
- Sentence splitting, which is where readability scorers actually fail — abbreviations, decimals, ellipses and quoted dialogue each turn one sentence into three and drag every metric with them
- The cost of an eager rewrite button: one "fix everything" click on a four-thousand-word draft is a couple of hundred API calls, and the retry loop makes it four hundred
- Your privacy assumption, the moment an autosave or a background lint starts posting the whole document to a model instead of the sentence you selected
- Your voice, if you accept every suggestion — which is what the tool is for and also what you notice a month later
- Nothing else. It is a text box with opinions about adverbs.
Is that you?
the verdict is a default, not a law
- You write for yourself and want the rules bent towards your own house style
- The AI rewrite is opt-in, per selection, and off by default
- The editor works with the network cable out, which is most of the reason to have your own
- Your documents are files on disk that outlive the app
- You are publishing the readability score as an authoritative number somebody else will act on — grade levels from different implementations are not comparable
- Your drafts are under embargo or NDA and the build sends whole documents to a third-party model without asking
- You are handing it to other people with your API key inside the client, which is a metered credential in a text box
If you build it anyway
the checklist, then the prompt that enforces it
- Use an existing readability library instead of implementing Flesch-Kincaid from a blog post, and name the formula and library version in the UI so the number means something.
- Fix sentence segmentation first. Every metric downstream is computed per sentence, and a naive split on full stops is already wrong in the first paragraph containing "e.g.".
- Keep the AI strictly opt-in and per selection. Send the highlighted sentence plus minimal context, never the document, and show the writer exactly what left the machine.
- Put the API key in the server environment rather than the client bundle, and cap spend per day even when the only user is you.
- Make the editor work offline with the highlighting intact. Local-only is the feature the paid product cannot sell you.
- Store documents as plain markdown on disk, saved on a timer, so the editor stays disposable and the writing does not.
I am building a readability editor: highlight hard sentences, adverbs, passive
voice and complex words, with an optional AI rewrite. It is a single-user tool
and should stay one. Build it in this order and argue if I reorder it.
1. Start with sentence segmentation, using a real library. Handle abbreviations,
decimals, ellipses, quotes and lists, and write tests for those cases before
anything is coloured — every metric depends on this one function.
2. Use an established readability package for the scores rather than writing the
formulae from memory. Print the formula name and library version beside the
number so it is comparable to something.
3. Make the highlighting rules data — a JSON list of patterns and word
substitutions I can edit — not conditionals buried in code.
4. Everything above must work with the network cable out. Prove it.
5. Documents are markdown files on disk with autosave. No database, no account,
no sync. If I ask for sync, tell me what it costs me in privacy first.
6. Only then add the AI rewrite, and only in this shape: I select a sentence, I
press a key, one request goes out with that sentence and at most a paragraph
of context. Never the whole document, never on a timer, never on save.
7. Show me in the UI exactly what text was sent and to which provider. If a
design you propose would send more than I selected, stop and say so.
8. The API key lives in a server-side environment variable. If any version puts
it in client code, refuse and explain why.
9. Add a hard daily request and token cap with a visible counter, and one
confirmation before any bulk operation over ten sentences.
10. Rewrites are suggestions in a diff view. Nothing overwrites my text without
an explicit accept, and undo restores the original exactly.
11. Out of scope on purpose: real grammar checking, collaboration, mobile. Point
me at LanguageTool for grammar, and note that the free Hemingway web editor
already does the highlighting for nothing.That one keeps you out of trouble. For the prompt that actually builds it, canivibecodeit.com has one.
their build prompt ↗Or don’t build it
the boring option, and the way back out
Worth saying first that the free Hemingway web editor already does the highlighting — $10/month buys the AI rewrites, the grammar checker and the desktop app. If what you want is rewrite quality, paying beats a weekend of prompt tuning. If what you want is the highlighting bent to your own house style, no subscription offers that and building is the right call.
$10/mo is cheaper than your weekend.
There is nothing to escape. The documents are markdown files, the rules are a JSON list you wrote, and deleting the app leaves you exactly where you started with the prose intact. That is the shape of a safe build: the artefact outlives the tool.
Mature open-source grammar and style checker with desktop and server components.
Python library implementing Flesch-Kincaid and the rest of the readability formulae, so you never write a syllable counter.
Questions
Will my readability score match Hemingway's?
Close, but not exactly, and nobody's matches anybody else's. Grade-level formulae depend on counting syllables and sentences, both of which are heuristics — hyphenation, abbreviations and dialogue punctuation each move the result. Use the number to compare two drafts of your own, not to win an argument with an editor.
Is sending my draft to an AI actually a risk?
For most writing, no — it is the same trade you make with any hosted editor. It becomes real in two cases: unpublished work under contract, and a build that quietly posts the whole document on every keystroke rather than the sentence you selected. The second is a design decision you can get right on day one and cannot notice later without reading your own network log.
Why is this SHIP IT when grammar tools usually aren't?
Because this one can be entirely local. The highlighting needs no server, no account and no upload, so the worst outcome is a repo you delete and a weekend you enjoyed. Add a hosted account holding other people's drafts and it becomes a different entry with a different verdict.
Every week, someone ships something they shouldn’t have.
New verdicts, the worst thing that landed in the trap, and the occasional incident report. No other email, ever.
A fiction brainstorming tool is prompts with a nice wrapper. Make it yours.
Twenty-five reports, and most of them are counting things. The one you can’t write is the grammar engine.
A grammar checker you run locally never has to be trusted with what you wrote.
last reviewed 2026-08-04 · verdict is editorial and unsponsored · shared entry data from canivibecodeit under MIT · not legal advice