Index of every page on this site: llms.txt
All skills
a2ui-project avatar

/natural-writing

@6bf4cc2
by A2UIa2ui-project/a2ui18k stars
1,430
Run it yourself

Contains well-defined rules for creating natural, accurate, and readable writing. Use this skill whenever authoring longer text, including reports, PR descriptions, READMEs, system designs, analysis documents, or general documentation.

Use this Skill: https://skilld.dev/gh/a2ui-project/a2ui/natural-writing

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

SKILL.md

≈63 tokens for metadata: the name and description. ≈2.2k when used: this file.

Description uses 3.2% of example budget

Before choosing a Skill, your Agent reads its name and description. All available Skills share that space.

  • A shorter description leaves more room for other Skills. This entry exceeds our 1% size suggestion.

In our Claude Code example, all Skill names and descriptions share 8,000 characters. This Skill uses ≈253 characters, or 3.2%.

The 1% threshold is a size suggestion. Longer descriptions can still fit.

Your model, settings, and other Skills decide how much text your Agent can read.

Example settings and source

The example uses a 200k-token context and default Claude Code settings. The count includes the name, description, separators, and when_to_use when present. Codex also counts local file paths.

Skit's source and limits: Codex 0.160.1, Claude Code 2.1.292.

Rules for natural writing

This document outlines strict rules to avoid common "AI-isms": stylistic and structural patterns that language models typically fall into. Follow these rules to produce content that is more understandable, and reads as natural, human-authored text.

The rules are numbered so that reviews can cite them, as in "rule 3.2".

1. Vocabulary & phrasing controls

1.1. The "banned" list

Avoid words that are statistically overrepresented in AI text. Use simpler, more direct alternatives.

Avoid verbs that are commonly overused, such as delve, underscore, highlight (as a verb), foster, cultivate, maximize, leverage, democratize, ensure, align with, resonate with, encompass, and bridge. Also avoid adverbs like seamlessly and extensively.

For nouns, avoid abstract uses of tapestry, landscape, realm, testament, interplay, synergy, cornerstone, hub, and ecosystem.

For adjectives, avoid puffery and vague descriptors like pivotal, crucial, vibrant, intricate, nuanced, unwavering, indelible, uncharted, rapidly evolving, transformative, breathtaking, nestled, dynamic, comprehensive, intuitive, holistic, robust, frictionless, scalable, and synergistic.

1.2. Avoid "copula" substitutions

Do not replace simple "is" or "are" verbs with flowery equivalents.

  • Instead of: "The library serves as a center for learning."
  • Write: "The library is a center for learning."
  • Instead of: "The statue stands as a monument to..."
  • Write: "The statue is a monument to..."

1.3. Eliminate "elegant variation"

Do not use synonyms just to avoid repeating a subject's name (e.g., "the eponymous character," "the titular protagonist," "the celebrated author"). It is acceptable to repeat the name or use pronouns naturally.

1.4. Prefer literal words to idioms

Choose the word that states the meaning directly over an idiom whose meaning depends on context. Idioms are harder to read for non-native speakers and for anyone skimming instructions.

  • Instead of: "Leave the version alone."
  • Write: "Leave the version untouched."

1.5. Name things in full for a reader without your context

Use the full name of a term when first mentioning it. Shorthand that was clear during development—such as an informal name for a part of the codebase may be meaningless to reviewers and future readers who lack that context.

  • Instead of: "The suites are read through _plain."
  • Write: "The conformance suites (the YAML test cases under conformance/) are read through _plain."

2. Content & tone

2.1. No "puffery" or forced significance

Do not inflate the importance of a topic with vague praise. If a subject is important, the facts should demonstrate it without help.

Avoid phrases like "serves as a testament to," "marking a pivotal moment," "underscoring the importance of," "leaving an indelible mark," or "shaping the landscape."

  • Instead of: "The founding of the institute marked a pivotal moment in the evolution of regional statistics, representing a significant shift toward independence."
  • Write: "The institute was founded in 1989 to collect regional statistics."

2.2. No superficial analysis

Avoid attaching "dangling" present-participle phrases that offer vague commentary.

Delete clauses starting with "highlighting," "emphasizing," "reflecting," "showcasing," or "demonstrating" if they just restate the obvious or add fluff.

  • Instead of: "The building uses blue glass, reflecting the region's natural beauty and symbolizing unity."
  • Write: "The building uses blue glass."

2.3. Avoid promotional language

Maintain a neutral tone. Avoid "advertisement" words like boasts, features (as a verb), offers, premier, leading, state-of-the-art, committed to, and dedicated to.

  • Instead of: "Nestled in the heart of the city, the hotel boasts a vibrant atmosphere."
  • Write: "The hotel is located in the city center."

2.4. No "challenges and future outlook" formula

Language models often end articles with a generic "Despite challenges... remains important" conclusion. Do not end with a summary paragraph starting with "Despite [X], [Subject] continues to..." or speculating on the future. End with the last fact.

  • Bad: "Despite facing economic hurdles, the company continues to thrive and remains a beacon of innovation."

2.5. No "title as proper noun" leads

Do not treat a descriptive article title (like a list or broad topic) as a proper noun in the first sentence.

  • Instead of: "The List of songs about Mexico is a curated compilation..."
  • Write: "This list contains songs about Mexico..."

2.6. No generic "see also" links

Do not populate "See Also" sections with broad, generic terms. Links must be directly relevant and specific to the subject.

  • Bad: Linking Financial technology in an article about a specific startup.
  • Good: Linking a competitor or specific related technology.

2.7. Attribution precision

Avoid vague "weasel words" like "Experts argue," "Observers have noted," or "Several sources indicate" unless you cite specific people immediately. Do not claim a subject interacts with a "broader" history or trend unless a source explicitly says so.

3. Sentence structure

3.1. No negative parallelism

Avoid sentences that structure a contrast unnecessarily.

  • Instead of: "It is not only a painting, but also a representation of..."
  • Instead of: "It is not just about X; it is about Y."
  • Write: "It is a painting that represents..."

3.2. No "rule of three"

Avoid listing exactly three adjectives or three noun phrases to sound "comprehensive."

  • Bad: "The event brings together marketers, engineers, and designers." (Unless those specific three groups are the only ones).
  • Bad: "It is bold, innovative, and unique."

3.3. No false ranges

Do not use "from X to Y" unless X and Y are endpoints of a logical scale (like time or size).

  • Instead of: "The book covers everything from biology to space travel." (These are just two random topics, not a range).
  • Write: "The book covers topics including biology and space travel."

4. Structure & formatting

4.1. Headers

Use sentence case for headers (e.g., "Early life," not "Early Life"). Do not use title case in headers.

4.2. Formatting avoidance

Do not use inline-header lists (such as * **Header:** Description...). Instead, use prose or simple lists.

Avoid excessive bolding of keywords, "key takeaways," or names in the body text, except for the first mention in the lead.

Use numbered lists (1., 2.) instead of bullets, so readers can refer to an item by its number, as in "see point 3" (except in changelogs, which use - bullets; see update-changelog). Do not use emojis (🚀, 🧠) as list markers.

Avoid creating tables for simple information that easily fits in a sentence.

Ensure markup is context-appropriate; do not use Markdown (like ##) in formats that do not support it (like Wikitext) unless it is explicitly converted.

4.3. Punctuation

Use straight quotes (", ') and straight apostrophes ('). Do not use curly or smart quotes (“, ’).

Use em dashes sparingly, as language models often overuse them for emphasis. Prefer commas or parentheses instead.

5. Citations & integrity

5.1. No hallucinations

Never generate a citation unless you are looking at the source. Do not invent URLs or DOIs. Do not assume a book exists or contains a specific fact without verification.

5.2. Link what you mention

Link everything the reader may want to open: files, functions, test and conformance cases, packages, issues, pull requests, and external pages. How precise a link should be depends on how long the text lives.

  1. In short-lived text, such as review comments, PR descriptions, and issues, link to the exact lines. Pin the link to a commit SHA rather than a branch, so it keeps pointing at the lines you meant after the branch moves. For a Markdown file, add ?plain=1 so a line anchor such as #L10-L20 works.
  2. In long-lived documents, such as READMEs, documentation, skills, and code comments, link to the file or its section, without line ranges. A line range in one long-lived document has to be updated every time the document it points to changes.
  3. Link only to what the reader can see at the link target. If your change is not pushed yet, link to the current version, or post the text after the push.
  4. Before posting, check that every link resolves.

6. Communication (chat context)

Avoid collaborative filler at the start of responses, such as "Certainly!", "Here is the information," or "I hope this helps." State the content directly.

Do not apologize for being an AI or mention knowledge cutoffs unless it is relevant to a specific, time-sensitive fact.

Do not preface responses with subject lines.

Keep edit summaries concise and informal, avoiding verbose paragraphs.

Source: SKILL.md on GitHub

No rule matched.

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

No alerts4mo3 checks · Risk SAFE
  • Gen Agent Trust Hub4mo

    The skill 'natural-writing' provides a set of stylistic rules and constraints designed to improve the quality and naturalness of AI-generated text. It contains no executable code, network operations, or malicious patterns.

  • Socket4mo

    No alerts

  • Snyk4mo

    Risk: LOW · No issues

Signed by skilld at 6bf4cc2. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub yesterday.

Activeupdated 2 weeks ago
metadata
{
  "internal": true
}

README badge

README badge for a2ui-project/a2ui/natural-writing