@borgesius/openclaw-style-guide

An OpenClaw context-engine plugin that injects writing style rules into every agent system prompt. Enforce consistent prose—Chicago, AP, MLA, or your own custom style—across all your AI outputs.

Why?

LLMs default to inconsistent, often corporate-sounding prose. They overuse em dashes, hedge constantly, and throw in phrases like "Certainly!" and "Great question!" that make your assistant sound like a chatbot.

This plugin fixes that by injecting clear, forceful style rules into every system prompt. Your agents write better because they have explicit rules to follow.

Installation

openclaw plugins install @borgesius/openclaw-style-guide

Or with npm:

npm install @borgesius/openclaw-style-guide

Then add to your openclaw.json:

{
  "plugins": {
    "entries": {
      "@borgesius/openclaw-style-guide": {}
    }
  }
}

Configuration

Configure via plugins.entries.@borgesius/openclaw-style-guide.config:

{
  "plugins": {
    "entries": {
      "@borgesius/openclaw-style-guide": {
        "config": {
          "preset": "chicago"
        }
      }
    }
  }
}

Presets

| Preset | Description | |--------|-------------| | chicago | Default. Chicago Manual of Style. Oxford comma, em dashes without spaces (used sparingly), spell out numbers 1–100. | | ap | AP Style. No Oxford comma, em dashes with spaces, spell out 1–9 only. | | mla | MLA Academic Style. Oxford comma, academic conventions. | | custom | Your own rules via the rules field. |

Custom Rules

Define your own style guide:

{
  "plugins": {
    "entries": {
      "@borgesius/openclaw-style-guide": {
        "config": {
          "preset": "custom",
          "rules": "Always use active voice. Keep sentences under 25 words. No jargon. Oxford comma required. Never start a sentence with 'I'."
        }
      }
    }
  }
}

What Gets Injected

When using chicago (the default), the plugin adds this to every system prompt:

## Writing Style: Chicago Manual of Style

You follow Chicago Manual of Style. These rules are non-negotiable:

**Em dashes:** No spaces around em dashes—like this, not like — this.
Use em dashes sparingly. Most of the time a colon, semicolon, or period
is the better choice.

**Oxford comma:** Always use the serial (Oxford) comma in lists of three
or more items: red, white, and blue.

**Quotation marks:** Double quotes for primary quotations. Single quotes
for quotations within quotations.

**Punctuation placement:** Periods and commas go inside quotation marks.
Colons and semicolons go outside.

**Numbers:** Spell out one through one hundred. Use numerals for 101 and
above. Spell out any number that begins a sentence.

**Voice:** No passive voice abuse. Prefer active constructions.

**Filler phrases to avoid:** Never write "certainly!", "great question!",
"I would be happy to", "absolutely!", or similar hollow corporate filler.
Just answer directly.

Token Cost

It's like 200–300 tokens per prompt. Basically nothing. Your conversations use thousands of tokens anyway, so this is <1% overhead for much better writing.

How well the model actually follows the rules depends on the model. Bigger models tend to be better at it.

How It Works

This is a context-engine plugin. It hooks into OpenClaw's before_prompt_build event and appends the style rules to the system context. The rules appear in every agent's system prompt, ensuring consistent style across all conversations.

TypeScript

The package exports types if you want to use the style blocks programmatically:

import { getStyleBlock, StyleGuideConfig, StylePreset } from '@borgesius/openclaw-style-guide';

const config: StyleGuideConfig = { preset: 'chicago' };
const styleBlock = getStyleBlock(config);

License

MIT