Research-backed writing skills, plugins, and MCP servers for drafting, editing, style, and translation.
| # | Name | Best for | Est. installs |
|---|---|---|---|
| 1 |  Internal Comms | Status updates and announcements | 120K |
| 2 |  Matt Pocock Writing Skills | Article drafting as a guided process | 90K |
| 3 | Doc Co-Authoring | Co-writing docs section by section | 85K |
| 4 |  Writing Guidelines | House style the agent actually follows | 50K |
| 5 |  Humanizer ZH | De-AI-ing Chinese prose | 45K |
| 6 | Baoyu Translate | Three-pass idiomatic translation | 40K |
| 7 | Humanizer | Stripping AI tells from English prose | 40K |
| 8 | Nature Skills | Academic papers and journal polish | 35K |
| 9 |  Lara Translate | Context-aware translation at scale | 30K |
| 10 | Stop Slop | Banning slop phrases outright | 25K |
| 11 |  DeepL | Team translation with glossaries | 25K |
| 12 |  Crowdin | Localization projects in Crowdin | 20K |
Anthropic’s official skill for internal communication: status updates, announcements, and team memos written in a clear, direct voice from the bullet points and context you give it.
Recurring team communication you write weekly anyway. It keeps the format consistent and cuts the drafting time; for substantial documents, use Doc Co-Authoring instead.
Four separately installable skills that form one long-form writing method: writing-fragments collects your raw material, writing-shape structures it, and writing-beats plans the argument beat by beat - shape and beats are alternative middle stages, pick one per piece - then edit-article rewrites the draft hard.
For articles and essays where one-shot generation produces generic output. The stages work individually - edit-article alone is a strong editor - but the full method is what turns notes into a piece with an actual argument.
Anthropic’s official co-writing skill. It aligns on purpose and audience first, drafts the document section by section, works your feedback in as it goes, and then tests whether a fresh reader would actually understand the result.
Workplace documents where structure carries the weight - specs, proposals, decision docs, reports. Less suited to short posts, where the process is more ceremony than help.
Vercel’s concrete prose rules packaged as a skill: sentence length, hedging, filler, structure. Instead of editing afterwards, it steers how the agent writes while it writes.
When you want a consistent standard across everything the agent produces rather than fixing each draft. Pairs naturally with Humanizer - guidelines steer during, Humanizer cleans after.
The Chinese-language adaptation of Humanizer. It targets the patterns that mark machine-written Chinese - translated-sounding constructions, filler transitions, over-formal register - and rewrites them into natural prose.
Any Chinese writing an agent produces that people will actually read. The gap between raw model output and natural prose is even wider in Chinese than in English, and this targets it directly.
A translation skill built on a three-step method: a literal pass, a critique of what reads unnaturally, then an idiomatic rewrite. Strongest for Chinese–English work, where the method originated.
Content you’ll publish, where a literal machine translation isn’t good enough. It costs more tokens than a one-shot translation and earns them on quality.
The original de-AI-ing skill. It works from a catalog of documented AI-writing signs - inflated transitions, empty emphasis, formulaic structure - and rewrites drafts so the patterns disappear rather than just get rephrased.
As a final pass over anything going out under your name. It edits what was written; if you want the agent to write differently in the first place, add Writing Guidelines alongside it.
A community suite of 18 academic-writing skills: paper structure, section drafting, polishing prose to journal register, reviewer responses, and figures. Each skill installs separately, and the count shown is the most-installed one (nature-figure) - adoption spreads across the suite rather than pooling in one skill.
Papers, theses, and grant writing - it encodes the conventions reviewers expect instead of leaving them to the model’s defaults. For general articles it over-formalizes; use Matt Pocock Writing Skills instead.
Translated’s official MCP server for its Lara translation platform: text translation, language detection, and context handling, served from a hosted endpoint with a local npm alternative from the same repository.
Volume translation through a dedicated engine rather than the model itself. The hosted endpoint sets up with a browser login, so there’s no local server to run.
A Lara account, authorized via browser OAuth; free, Pro, and Team tiers are documented.
A blunt instrument: hard bans on the words and constructions that mark AI slop. No rewriting philosophy, just a list the agent is not allowed to touch.
When Humanizer’s full rewrite is more than you want and you just need the worst offenders gone. Cheap to run, easy to keep on permanently.
DeepL’s official MCP server, run locally from npm and backed by the DeepL API: text and document translation, rephrasing, language operations, and your organization’s glossaries.
Teams that already standardize on DeepL and need terminology to stay consistent across translations. For one-off translation, a skill like Baoyu Translate costs nothing.
A DeepL API key and Node.js 18+ - the server runs on your machine via npm.
Crowdin’s official hosted MCP server for localization operations: projects, source strings, translations, and terminology, worked on from inside the agent.
Ongoing product localization - many languages, many strings, continuous updates. Not for translating a single document; use DeepL or a translation skill for that.
A Crowdin account with a localization project.
What are you writing?
→ Matt Pocock Writing Skills - articles and essays that need an actual argument, not a generic draft
→ Doc Co-Authoring - specs, proposals, and decision docs, built section by section
→ Internal Comms - the status updates and announcements you write every week
→ Nature Skills - papers and theses held to journal conventions
Want prose that doesn’t read like AI?
→ Writing Guidelines - steers the style as it writes, so drafts start clean
→ Humanizer - rewrites English drafts against the documented AI tells
→ Humanizer ZH - the same for Chinese, where the gap is wider
→ Stop Slop - a cheap, always-on ban on the worst phrases
Guidelines steer during, Humanizer cleans after - they stack.
No official Grammarly route yet - these target AI tells and house style, not comma-level grammar.
Translating something worth getting right?
→ Baoyu Translate - three-pass idiomatic quality, no account needed
→ Lara Translate - volume through a hosted engine, set up with a browser login
→ DeepL - when a team needs glossaries and consistent terminology
→ Crowdin - ongoing product localization, not a one-off document
The skill needs no account; the MCP servers pay off on volume and shared terminology.
