Claude Writing Skills, Ranked by Verdict

Humanizer (72, Recommended with caveats) is the best Claude writing skill we have reviewed. None of the five earns Recommended, and four name an invented or altered fact as their worst failure.

Claude writing skills are SKILL.md folders that change how Claude drafts and edits prose. Claude's Help Center defines skills as "folders of instructions, scripts, and resources that Claude loads dynamically." This page ranks writing skills, not Claude models. Our primer explains what Agent Skills are and how they load.

What are the best Claude writing skills?

Humanizer (72, Recommended with caveats) is the best Claude writing skill we have reviewed. Anthropic's Doc Co-Authoring (67) and Internal Comms (64) follow. None of the five Claude writing skills earns Recommended.

Marketing copy skills, such as Copywriting and Copy Editing, are not ranked here.

How were these Claude writing skills ranked?

The list is sorted by score. The score weights each skill's worst failure. For four of these five writing skills, that failure is an invented or altered fact. Seven universal measures sit beside measures built for the writing job. Our methodology page shows how we score and grade skills.

How does this ranking differ from star-ranked lists?

Analytics Vidhya ranked five writing repos by GitHub stars on 2026-08-05. A dev.to listicle ranked skills by its author's own five-axis test. We rank by verdict at a pinned commit. Beautiful Prose has 57 stars and scores 62. Content Research Writer sits in a 76,196-star repo and scores 58.

Which Claude writing skill fits which writing job?

Each Claude writing skill fits one writing job. Humanizer cleans AI-drafted text, Doc Co-Authoring interviews you through a spec, and Internal Comms formats team updates. Beautiful Prose sets a terse voice; Content Research Writer structures articles.

Humanizer: cleaning text a model drafted

Humanizer rewrites AI-sounding prose against 26 patterns drawn from Wikipedia's "Signs of AI writing." Judges preferred its rewrite 16 of 16 times in a blind study (#229). A GPT-5.4 replay on 18 drafts cut a slop score from 76.3 to 35.4. That replay kept 17 of 18 details (#250). Its own worked examples in §12, §18 and §25 add or change facts. Fix requests #233 and #306 were closed unchanged. The file measures 8,087 tokens per load. File mode overwrites the named file without asking (#238).

Doc Co-Authoring: specs, RFCs and proposals with a human present

Doc Co-Authoring is Anthropic's three-stage workflow: interview, draft section by section, then test with a fresh reader. In one hands-on report (koldovsky PR #25), that reader test caught a factual contradiction and six other clarity problems. The file measures 3,954 tokens per load. Delivery breaks without a human present. In issue #1718, 0 of 15 engaged runs wrote the requested file in single-turn use. Fix PR #1759 was still open on 2026-10-21. Use it in an interactive session where you answer its questions, as the Doc Co-Authoring verdict explains.

Internal Comms: 3P updates, newsletters and FAQ roundups

Internal Comms is a light template pack from Anthropic. It loads about 378 tokens and ships guides for 3P updates, a company newsletter and an FAQ roundup. The 3P format (Progress, Plans, Problems) is strict and easy to check. Four of the seven advertised types have no guide of their own. Status reports, incident notes and leadership updates fall back to nine lines of generic advice. No rule forbids invented metrics. The content has stayed unchanged since December 2025.

Beautiful Prose: fresh drafts in a hard, declarative voice

Beautiful Prose is a style contract that bans em dashes, pivot phrases and filler. Register, density, heat and length tags let you set the voice. It measures 1,227 tokens per load and holds no code. Nothing in it tells the model to keep facts, figures or qualifiers when it rewrites. We found no licence file, and the last commit landed on 2025-12-30. Pick it for new drafts you will fact-check yourself.

Content Research Writer: outlines and section feedback for long articles

Content Research Writer runs an eight-step co-writing routine from outline to final review. Its templates cover hooks, section feedback and a pre-publish checklist. Its research examples show invented statistics, a fictional MIT quote and placeholder citations. It names no search tool and has no verification step. Tessl rates the file 51/100 for verbosity. Supply your own sources and treat it as a structure coach.

Which Claude writing skills should you avoid?

No Claude writing skills are Not Recommended. Avoid using Content Research Writer (58) for research, because its examples model invented statistics and citations. Four of five name an invented or altered fact as their worst failure, so check every claim.

Each verdict scores the parameter tied to that skill's worst failure. These five scores show where each writing skill breaks:

1. Citation integrity scores 35 on Content Research Writer, the lowest parameter on this page.
2. Fact and qualifier preservation scores 45 on Beautiful Prose.
3. Fact preservation scores 55 on Humanizer, despite its rules against invention.
4. Privacy and audience control scores 45 on Internal Comms.
5. Artifact delivery when interaction is impossible scores 25 on Doc Co-Authoring.

Four of five skills name an invented or altered fact as their worst failure. None ships an automated fact check. Every skill holds safety grade A or B and none needs network access. The risk with these skills is accuracy, not security. Our guide covers how to vet a skill before installing it.

Which Claude writing skills remove AI tells and make text sound human?

Humanizer (72) is the Claude writing skill that removes AI tells, backed by two third-party measurements. Beautiful Prose (62) bans similar tics in fresh drafts at under a sixth of the token load. Nothing measures its effect.

Humanizer works on finished text and returns a draft, a list of remaining patterns and a final rewrite. Beautiful Prose works earlier, as a rule set Claude follows while it writes. Nobody has published a measured before-and-after for Beautiful Prose.

Neither skill is built to pass AI detectors. Humanizer's README calls detector evasion a non-goal. Issues #263 and #281 report Pangram and checker failures after rewrites.

The pattern count differs by source. Analytics Vidhya says Humanizer checks "33 patterns." The README at commit 225a6f3 says 26, and it still said 26 on 2026-10-21. We use the README figure. Our page covers our full Humanizer verdict and its alternatives.

Three anti-slop skills are not yet evaluated: stop-slop, avoid-ai-writing and ZeroSlop. Analytics Vidhya listed stop-slop at 15,199 stars on 2026-08-05.

Is there a Claude writing skill for blog posts, essays and business documents?

Yes, three: Content Research Writer (58) outlines and critiques long articles. Doc Co-Authoring (67) drafts specs and proposals; Internal Comms (64) formats team updates. No reviewed Claude writing skill targets essays or whitepapers.

Each of the three covers a different document type:

• Blog posts and long articles: Content Research Writer gives outlines, research prompts, hook options and section feedback. Our estimate for a 1,500-word article is 60,000 input and 10,000 output tokens. skills.sh lists 7.7K installs.
• Specs, design docs and proposals: Doc Co-Authoring interviews you, then drafts section by section. Our estimate for a 2,000-word design doc is 120,000 input tokens over several rounds.
• Team updates: Internal Comms covers the 3P format, a 20 to 25 bullet company newsletter and an FAQ roundup.

Essays get no dedicated skill. Beautiful Prose comes closest for fresh essay drafts in a set register. Our reviews point to one workflow for any of these jobs. Draft with the skill that fits, run Humanizer on the result, then diff the numbers against your source.

claude-blog (AgriciDaniel) is not yet evaluated. The same author publishes Claude SEO, a separate skill we have reviewed.

Newsletters, reports and whitepapers

Internal Comms covers internal newsletters only. Substack newsletters, whitepapers, sales email and grant proposals have no reviewed skill yet. Doc Co-Authoring is the nearest fit for reports. Its fresh-reader test checks a draft for gaps before others read it.

Can a Claude writing skill learn your voice, or should you build your own?

Yes, partly. Humanizer can rewrite in your voice from a writing sample, and Beautiful Prose offers fixed register tags. For a reusable voice, build your own Claude writing skill from samples of your published writing.

Two reviewed skills handle voice today. Humanizer lists "Rewrite in the user's own voice from a writing sample" as a secondary use case. Its voice and restraint parameter scores 65. Users report over-correction of human text, and the maintainer declined a density pre-check request (#93). Beautiful Prose offers four registers plus density, heat and length tags. Those tags set a voice, but not yours.

A skill of your own carries your voice into every session. Four steps build one:

1. Collect samples of your published writing; the aiblewmymind guide recommends 5 to 10 pieces.
2. Ask Claude to describe your tone, rhythm, sentence length and banned phrases.
3. Save those rules as a SKILL.md with a description that says when to use it.
4. Test the skill on a new draft, then revise the rules where it misses.

Three published guides show the approach. aiblewmymind builds a Voice DNA skill and an audience-profile skill. Allen Yang describes a personal voice skill on LinkedIn. Superpath turns reviewer personas into a feedback skill.

Claude can draft the skill file itself. Anthropic's Skill Creator automates that and scores Conditional at 57, as Anthropic's Skill Creator verdict explains. Our guide shows how to write your own Agent Skill from scratch. Published "voice DNA" and "write like me" packages are not yet evaluated.

Which writing jobs have no reviewed Claude skill yet?

Résumés, cover letters, fiction and translation have no reviewed Claude writing skill yet. For academic writing, the nearest fit is Scientific Agent Skills, a research library with reference checks. This hub ranks none of them.

Résumés and cover letters

Résumé writing is the largest writing job we have not yet reviewed. No verdict covers it. tailored-resume-generator showed 9.7K skills.sh installs when we checked and is not yet evaluated. Our Beautiful Prose verdict lists résumés as a poor fit, because its bans on lists get in the way. Use a checklist for skills we have not reviewed before you install one.

Academic and research writing

Scientific Agent Skills is a research library, so it is not ranked here. AI Research Skills (Orchestra) is another research library. For thesis and research-paper writing, these two are the nearest fits, not a ranked writing skill.

Fiction, scripts and translation

Fiction and script skills are not yet evaluated, including oh-story-claudecode. One adopter uses Beautiful Prose for game narrative text, with a custom lint in CI (Rutilusaduro/MDSim). Translation skills are not yet evaluated either.

Do Claude writing skills work in Codex, Cursor and other agents?

Yes in Claude Code, rarely elsewhere: three of five Claude writing skills have verified Claude Code runs. Humanizer is verified in Claude Desktop and Cowork too. Beautiful Prose alone is verified in Cursor; none is verified in Codex.

Internal Comms and Content Research Writer have no verified agent at all. Doc Co-Authoring names claude.ai tools such as create_file and str_replace. Other harnesses lack those tools, so its steps do not map one to one. Humanizer's Codex, Cursor and other entries are publisher claims, with no verified run. Our agent hub lists Claude Code skills, ranked across every job. A wider list covers the best Claude skills across every job in the Claude app.

How do you install a Claude writing skill?

Install a Claude writing skill through Claude Code's plugin marketplace, the skills CLI, or a folder copy into ~/.claude/skills/. In the Claude app, upload its ZIP under Customize > Skills. Claude loads it when a request matches.

Humanizer's README gives the commands below. Other skills follow the same routes with their own repo names.

1. Open the verdict and check that the reviewed version matches the one you install.
2. Add the plugin in Claude Code with /plugin marketplace add blader/humanizer, then run /plugin install humanizer@humanizer. The README says this needs Claude Code 2.1.142 or later.
3. Run npx skills add blader/humanizer --global --agent claude-code to use the skills CLI instead.
4. Copy the skill folder into ~/.claude/skills/ for every project, or .claude/skills/ for one.
5. Upload the ZIP in the Claude app under Customize > Skills, with code execution turned on.

Humanizer needed two Claude Desktop fixes: v2.10.2 for the loader (#202) and v2.11.1 for ZIP symlinks (#224). Our setup guide shows how to install skills in Claude Code. A second guide explains how to turn on skills in the Claude app. A third covers why a skill does not trigger.

Where are the best Claude writing skills on GitHub?

The five reviewed Claude writing skills live in four public GitHub repos: anthropics/skills, ComposioHQ/awesome-claude-skills, blader/humanizer and SHADOWPR0/beautiful_prose. Stars measure popularity, not quality. The top-scoring skill sits in neither of the two largest repos.

Content Research Writer's folder has not changed since our review at commit be2a406. None of the five SKILL.md files changed between our review and 2026-10-21.

Common questions

Is there a writing skill for Claude?

Yes: we have reviewed five Claude writing skills. Humanizer leads at 72, Recommended with caveats.

How do I write a skill for Claude?

Write a SKILL.md with a name, description and instructions, per our guide to create an Agent Skill. A second page explains how Claude picks a skill at run time.

Can Claude write its own skills?

Yes: ask Claude to draft one, or use Anthropic's Skill Creator, which scores Conditional at 57. Its full Skill Creator review lists the trade-offs.

Are Claude writing skills free?

Yes: all five are public on GitHub at no cost. We found no licence for Beautiful Prose or the two Anthropic folders.

What does the community report about Claude writing skills?

Two Hacker News comments on Humanizer and three GitHub threads (PR #233, issues #306 and #1718) hold the evidence.

Do writing skills make Claude text undetectable?

No: Humanizer's README calls detector evasion a non-goal. One Turnitin report (#252) is anecdotal.

Is there a Claude skill for UX writing?

Yes: Anthropic's Design Plugin includes a UX copy skill and scores Conditional at 66. It ranks among UI/UX and frontend design skills.

Every ranked list on the site starts from all best Agent Skills, ranked.