skill

anti-ai-writing-skill

AI skill that removes the machine-made tells from your writing: vague claims, 'not X but Y' reframes, hype words, fake metaphors. Works with Claude Code, Cowork, Codex, ChatGPT, Grok.

swayamdhawan/anti-ai-writing-skillcommunitylistedunreviewed

Use skill

Set up

Skills are installed by placing their folder where your agent loads skills.

Download ZIP from GitHub Open the repository

The download comes from github.com/swayamdhawan/anti-ai-writing-skill itself, not from our server, so what you get is exactly what its author published.

Install instructions come from the README

Skills do not use an MCP config. Follow the steps in the README below.

README

synced Oct 22, 2026

anti-ai-writing

Stop sounding like AI. Sound like a specific person with something to say.

AI writing fails the same few ways: vague claims, "It's not X, it's Y" reframes, fake-deep metaphors, hype words, and engagement bait. Readers spot it in one line and scroll. anti-ai-writing is an AI skill that runs as the final filter on anything you write and removes the machine-made parts.

It does not just ban words. Blocklists rot. The skill goes after the real problem: specificity and voice.

Works with Claude Code, Claude Cowork, Claude desktop and claude.ai, OpenAI Codex, ChatGPT, Grok, and any other AI chat. Built by Creatofy.

See it work

Before

It's not about the prompt. It's about the context.

After

The model gets about 3x better when you paste the previous chat into the system prompt.

Before

In today's fast-paced digital landscape, it's important to note that leveraging AI can unlock transformative growth for your business.

After

I run a four-person agency. Last month AI cut our proposal time from 3 hours to 40 minutes. Here is exactly what I changed.

(Sample facts are made up for the demo. The skill never invents numbers. If it lacks a specific, it asks you or leaves a [bracket].)

What it catches

The 5 diseases

Name the problem and the fix is obvious.

Vagueness compression: describes a category, not a thing.
Significance inflation: treats normal facts like turning points.
Hedged confidence: has a position but won't commit.
Rhythmic flatness: every sentence the same length.
Borrowed authority: sounds like a consulting deck or a thinkfluencer.

The specificity ladder

The highest-leverage rule. Aim for level 3 or 4.

Level	Example
Vague	"The company faced challenges."
Specific	"The company had cashflow issues."
Concrete	"The company missed payroll twice in six months."
Lived	"Payroll bounced in March and again in August. The CFO found out from Slack."

The #1 tell: negative parallelism

"It's not X. It's Y." and every cousin: Less X, more Y. / Stop thinking X, start thinking Y. / Most people think X, but... The fix is to delete the rejected half and write the other half as a direct claim with specifics.

It also knows when a contrast is earned (the second half is concrete and the piece pays it off) versus hollow (vague significance hiding the lack of a point). Good hooks keep their contrast. Empty ones get cut.

Metaphor control: writes literally by default. Banned families (flywheel, DNA, iceberg, leaky bucket) and banned verbs (baked in, bolted on, unpacked, woven).
Vocab and phrase tells: delve, leverage, seamless, game-changer, "Let's dive in", "Let that sink in", "Furthermore", and more.
Formatting tells: em dashes, hashtag and emoji spam, walls of bullets, the rule-of-three reflex.
Other tells: false ranges, elegant variation, participle fake-depth, symmetric slogans, throat-clearing first sentences.

Two modes

Mode	You say	You get
Write	"run anti-ai-writing on this" + your draft	A direct, specific rewrite in your voice
Audit	"audit this, does it sound like AI?"	Each tell flagged with the exact line, then a rewrite

ANTI-AI AUDIT:
  Negative parallelism: FLAG - "They're not just employees. They're the DNA of everything we do."
  Metaphor:             FLAG - "DNA of everything" (banned family) - say what they actually shape
  Specificity:          FLAG - no names, numbers, or lived detail
  Rewrite: "Maya in support rewrote our refund policy in April. Complaints dropped by half."

Built-in judgment

Priority order: Accurate > Clear > Specific > Voiced > Stylish. A boring true sentence beats an elegant vague one.
Spirit beats letter. If a rule makes the sentence worse, it breaks the rule.
Written vs spoken: full pass on captions, carousels, newsletters, threads, LinkedIn, and long-form. A lighter subset for reel and video scripts, where formatting tells don't exist.

Install

Pick your tool.

Claude Code, Cowork, and other Claude CLIs (GitHub CLI)

gh repo clone swayamdhawan/anti-ai-writing-skill
cd anti-ai-writing-skill && ./install.sh claude

Or one line, no script:

gh repo clone swayamdhawan/anti-ai-writing-skill /tmp/anti-ai-writing-skill && mkdir -p ~/.claude/skills && cp -R /tmp/anti-ai-writing-skill/anti-ai-writing ~/.claude/skills/anti-ai-writing

For one project only, run ./install.sh claude --project inside that project. Restart Claude Code if it was already running.

Claude desktop, Cowork, and claude.ai (upload)

Download the ready-made file with GitHub CLI:

gh release download --repo swayamdhawan/anti-ai-writing-skill --pattern 'anti-ai-writing.skill'

Then open Settings > Capabilities > Skills (on some plans Customize > Skills), click Upload skill, pick anti-ai-writing.skill, and make sure it is toggled on.

OpenAI Codex CLI

gh repo clone swayamdhawan/anti-ai-writing-skill
cd anti-ai-writing-skill && ./install.sh codex

This copies the skill to ~/.codex/skills/anti-ai-writing. Restart Codex. If your Codex version does not load skills, paste prompts/anti-ai-writing-prompt.md into your AGENTS.md instead.

Install for both at once with ./install.sh all.

ChatGPT, Grok, Gemini, and any chat AI

These don't load skill folders, so use the paste-in version. It has the same rules.

gh repo clone swayamdhawan/anti-ai-writing-skill /tmp/anti-ai-writing-skill && pbcopy < /tmp/anti-ai-writing-skill/prompts/anti-ai-writing-prompt.md

(pbcopy is macOS. On Linux use xclip -selection clipboard <, on Windows use clip <.)

Then pick one:

ChatGPT: paste into a new chat, a Custom GPT's instructions, a Project's instructions, or Settings > Personalization > Custom instructions.
Grok: paste into a chat, or into your custom instructions in settings.
Gemini: paste into a chat, or create a Gem with it as the instructions.

Then send your draft with "run anti-ai-writing on this".

Plain terminal (no install)

gh repo clone swayamdhawan/anti-ai-writing-skill && cat anti-ai-writing-skill/anti-ai-writing/SKILL.md

How to use it

In Claude Code or Cowork:

/anti-ai-writing
In today's fast-paced digital landscape, it's important to note that leveraging AI can unlock...

Or just talk to it:

"make this sound less like AI"
"audit my LinkedIn post"
"this feels generic, fix it"
"run the anti-AI pass on this newsletter"

It triggers automatically as the final filter on written content, and whenever text feels generic, hollow, over-formatted, or "sounds like AI".

Pairs well with

dumbify: lowers reading level so content is easy to follow. Run dumbify for clarity, then anti-ai-writing for voice.

Repo layout

anti-ai-writing/
  SKILL.md                       the skill (this is what agents load)
prompts/
  anti-ai-writing-prompt.md      paste-in version for ChatGPT, Grok, Gemini
examples/
  before-after.md                rewrites and an audit example
install.sh                       installer for Claude and Codex
LICENSE                          MIT

License

MIT. Use it, fork it, remix it. A star on the repo helps others find it.

Made by Swayam Dhawan at Creatofy.