Skip to content
Development
Skill

/writing-guidelines

This skill should be used when the user asks to "review writing", "check documentation style", "audit interface copy", or "apply writing guidelines".

From plugin
claude-codex-settings
1.1k78 skills4 agents5 commands5 MCP
Install
$ npx -y skills add fcakyon/claude-codex-settings --skill writing-guidelines --agent claude-code

How it fires

How this skill gets triggered: by you, by Claude, or both.

  • Fires itselfAuto-invocation. Claude auto-loads it when your prompt matches the work.Auto-invocation is when the right skill fires by itself at the right moment, driven by a FLOW.md router and a hook, instead of you invoking it by name. It is the difference between a skill being installed and a skill actually getting used.Read the full definition →
  • You can call itInvoke it directly when you want it.
  • Slash command/writing-guidelines

Context preview

The summary Claude sees to decide when to auto-load this skill.

This skill should be used when the user asks to "review writing", "check documentation style", "audit interface copy", or "apply writing guidelines".

SKILL.md

writing-guidelines.SKILL.md
name: writing-guidelines
description: This skill should be used when the user asks to "review writing", "check documentation style", "audit interface copy", or "apply writing guidelines".
metadata:
  argument-hint: <file-or-pattern>
license: MIT

Writing Guidelines

Review the files or patterns provided by the user for compliance.

Read files, check against rules below. Output concise but comprehensive: sacrifice grammar for brevity. High signal-to-noise.

Rules

Planning & content type

  • Every page has a plan (overview, goal, audience, content plan, open questions) referenced or linked
  • Content type declared in `meta.contentType`: `Tutorial`, `How-to`, `Reference`, `Conceptual`, `Troubleshooting`, or `Landing`
  • Title is user-shaped (the user's question), not feature-shaped (the engineer's name)
  • Page does one job: tutorial OR how-to OR reference, not three at once
  • Goal is verb-driven (Bloom's taxonomy): "configure", "explain", "debug" (testable)
  • Multi-audience pages: short shared opener, then technical subsections

Voice & tone

  • Active voice. Mental test: append "by monkeys". If the sentence parses, rewrite
  • Direct address: `you`, never `the user` or `one can`
  • Imperative for steps: "Click **Add Project**", not "You will need to click **Add Project**"
  • Sentences under 20 words target
  • Contractions encouraged (`you'll`, `it's`) for warmth
  • Present tense unless describing future behavior
  • Limit `we`: only for deliberate organization actions ("we recommend", "we deprecated"), never as a stand-in for "you"
  • No rhetorical questions (sounds like marketing)
  • Second-read test: read each sentence once at speech pace; if you re-read to parse it, name the subject, the action, and the consequence (kill metaphor verbs and pronouns reaching back several sentences)

Banned words

  • `easy`, `simple`, `quick`: puts pressure on the reader and reads as marketing; replace with concrete description ("one command", "default settings", "most projects don't need this")
  • `very`, `just`, `really`: filler; cut or rewrite

Concision

  • Earn every detail: cut a number, name, or implementation detail if a more general phrasing wouldn't change the reader's understanding or action
  • Weasel words: replace vague qualifiers (`significantly`, `many`, `often`, `typically`, `generally`) with a specific number or claim
  • Vague quantifiers: no `near-zero`, `sub-second`, `most requests`; give the figure and cite it (`99.37% of requests see zero cold starts`)
  • Filler/metaphor verbs: name the action instead of reaching for cadence (`moves through`, `lands`, `carries`, `hits` → the literal step)

AI-generated tells (flag these)

  • Summary-style transitions: never open a paragraph by recapping the last one (`With this setup complete…`, `Now that we've explored…`); pivot straight to the next point (`In practice…`, `The catch is…`)
  • Stop-start sentences: don't split one dependent idea into choppy fragments (`Previously this was manual. Now it's automatic. This saves time.` → one sentence); short sentences for emphasis are fine
  • Spec-sheet voice: rewrite sentences that read like a system reading a datasheet (`provides`, `is configurable`, `is explicitly labeled`)
  • Cold-open paragraphs: a body paragraph whose first sentence works as a standalone heading has no antecedent; carry the prior subject forward (`Because…`, `Once…`)
  • Personified artifacts: machines don't perform human-physical actions (`hand the browser a URL` → `the browser fetches the URL`; `the token holds…` → `the token is stored…`)
  • Reused framing: the angle must come from this page, not a template (`The question most teams face is whether…`)

Tone, by content type

  • **Tutorial**: warm, encouraging, predictable structure, no traps
  • **How-to**: terse, direct (reader is mid-task)
  • **Reference**: neutral, exhaustive, quotable
  • **Conceptual**: explain like the reader will teach it back; examples and analogies welcome
  • **Troubleshooting**: empathetic but not apologetic; acknowledge then fix

Headings

  • Sentence case for page headings (`H1` `H2` `H3`): "Configure environment variables", not "Configure Environment Variables"
  • Title case for nav labels: "Configuring Environment Variables"
  • `meta.title` becomes the `H1`; `meta.navLabel` becomes the sidebar entry
  • Subheadings descriptive, not cute: "Caveats when self-hosting on Cloudflare", not "Caveats"
  • Reader should be able to guess section content from the heading alone

Structure

  • Every page opens with a one-paragraph TL;DR of what the page covers
  • Every major section opens with a summary sentence
  • Acronyms spelled out on first use: "Content Security Policy (CSP) blocks inline scripts"
  • Define every term the first time you use it (link to its conceptual page)
  • Reference docs organized by surface; education docs organized by reader task
  • Keep paragraphs to 2 to 4 sentences; split anything longer or covering two ideas

Lists

  • Three or more list-shaped items in a paragraph: convert to a list
  • Bulleted for unordered; numbered for ordered (lifecycles, sequential steps)
  • Always introduce a list with a colon
  • No periods at the end of list items unless they are full sentences
  • Bold/description format: `- **Term**: description here` (colon after bold term)

Code

  • Code blocks need a language tag for syntax highlighting
  • TypeScript is the default for new code unless the surface is genuinely language-agnostic
  • Multi-step flows wrapped in `<Steps/>` so structure is visible
  • Highlight load-bearing lines: `` ```typescript {8-12,23-37} ``
  • ≤80 columns per line in snippets
  • ≤25 lines per snippet; split longer blocks with prose
  • Omit defaults; don't repeat variable definitions, use shared var
  • Minimal comments in code blocks; prefer prose explanation
  • Explain what every code block does in prose (don't drop and run)
  • Don't reference full example files at the end of guides ("See `train.py`"); the guide is the deliverable

Placehold

Read more
Ships withclaude-codex-settings

Battle-tested Claude Code, OpenAI Codex, Cursor configs, plugins, hooks and agents with Kimi, MiniMax and GLM API support.

Get the whole plugin

Other skills on claude-codex-settings.