DEV Community

Nova
Nova

Posted on

Build a Personal Writing Linter with AI: Style Specs + Self-Checks

#beginners

Most people use LLMs for writing like this:

“Rewrite this to sound better.”

And then they wonder why the output feels:

  • generic
  • too long
  • weirdly confident
  • inconsistent with their voice

Here’s a better approach that works extremely well for developers:

Treat writing like code.

You already know how to get consistent code quality:

  • a style guide
  • a linter
  • a review checklist

You can build the same thing for writing with a simple prompt + a repeatable workflow.

Step 1: write your “style spec” (copy/paste)

Create a short style spec you can reuse. Keep it explicit.

STYLE SPEC
- Voice: knowledgeable, casual, senior dev. No hype.
- Sentences: mostly short. Avoid long chains of commas.
- Structure: strong headings, bullets, concrete examples.
- Avoid: buzzwords, vague claims, "revolutionary", "game-changing".
- Prefer: active voice, specific nouns, numbers.
- Audience: developers.
- Goal: teach something practical.
Enter fullscreen mode Exit fullscreen mode

This alone fixes 50% of “LLM writing” problems.

Step 2: create a writing checklist (the linter rules)

These are my default checks:

  • Does the intro explain the problem in 2–3 sentences?
  • Are there copy/paste examples?
  • Are claims backed by specifics?
  • Is the conclusion actionable?
  • Are there any “assistant-y” phrases? (remove them)

Now we turn that into a prompt.

Step 3: the “AI linter” prompt

Paste your draft, get a structured review.

You are my writing linter.

STYLE SPEC:
<PASTE STYLE SPEC>

TASK:
Review the draft and output:
1) A list of issues (bullet list) grouped by: Clarity, Structure, Tone, Specificity.
2) A suggested rewrite of the intro (max 120 words).
3) A list of 5 concrete edits (each as "Before -> After").

RULES:
- Do not rewrite the whole article.
- Be direct. No compliments.
- If something is vague, propose a specific replacement.

DRAFT:
<PASTE DRAFT>
Enter fullscreen mode Exit fullscreen mode

This keeps the model in “review mode” instead of “rewrite everything mode”.

Step 4: apply edits with a second prompt (safely)

Once you have review feedback, apply it with constraints.

Apply the following edits to the draft.

Rules:
- Keep meaning.
- Keep headings.
- Do not add new sections.
- Keep total length within +/- 10%.

Draft:
<PASTE>

Edits to apply:
<PASTE LINTER OUTPUT>
Enter fullscreen mode Exit fullscreen mode

This two-step flow produces writing that feels like you, not like a generic content bot.

Automate it: run the linter from the CLI

Here’s a minimal Node.js script that:

  • reads a markdown file
  • prints the linter prompt with your style spec included 
// lint-prompt.js
import fs from "node:fs";

const style = fs.readFileSync("style-spec.txt", "utf8");
const draft = fs.readFileSync(process.argv[2], "utf8");

const prompt = `You are my writing linter.

STYLE SPEC:\n${style}

TASK:
Review the draft and output:
1) A list of issues grouped by: Clarity, Structure, Tone, Specificity.
2) A suggested rewrite of the intro (max 120 words).
3) 5 concrete edits (Before -> After).

RULES:
- Do not rewrite the whole article.
- Be direct. No compliments.

DRAFT:\n${draft}
`;

console.log(prompt);
Enter fullscreen mode Exit fullscreen mode

Usage:

node lint-prompt.js article.md | pbcopy
Enter fullscreen mode Exit fullscreen mode

You can do the same in Python, Makefiles, or whatever you already use.

The “final pass” prompt: catch last-mile problems

Before publishing, I run a final pass with a short checklist:

Final pass.

Check the draft for:
- unclear sentences
- missing context
- passive voice
- repeated phrases
- paragraphs longer than 5 lines

Output ONLY a bullet list of fixes. No rewrites.

Draft:
<PASTE>
Enter fullscreen mode Exit fullscreen mode

It’s boring. It works.

Why this works

You’re not asking the model to “be creative.”

You’re asking it to:

  • follow a spec
  • run a checklist
  • propose diffs

That’s exactly the kind of task LLMs are good at.

Want the templates I actually use?

I keep a curated set of prompt templates for dev + productivity workflows in my Prompt Engineering Cheatsheet.

Top comments (0)