Resource · Free skill for Claude

Cervantes
Social Media

LinkedIn has been publishing the same post for years. Only the industry and the photograph change. Cervantes detects 82 habits of AI writing in Spanish and helps remove them before you publish another one.

Version 1.7 · 22 KB · No registration · MIT License

01

The fatigue

Why I made this

I was tired of reading posts written from the same recipe. An “uncomfortable truth” in the first line, three perfectly measured lessons, an arrow or two, and a question at the end designed to manufacture comments.

Quite often there was a good idea and an interesting person underneath. The template had swallowed both. I started collecting the recurring constructions and ended up with a catalogue of 82 patterns.

Cervantes turns that catalogue into instructions for Claude. It makes Claude look for one idea, an opinion and something concrete before writing. A separate second pass then hunts for any habits that slipped through.

02

The collection

The usual suspects

There are 82 patterns, grouped by structure, punctuation, vocabulary, anglicisms, hooks, figures, formatting and endings. These Spanish specimens are remarkably punctual:

The contrast

«No es suerte. Es disciplina.»

The reveal

«Y la razón es simple: nadie escucha.»

The consultant

«Una palanca fundamental para potenciar el ecosistema.»

The ending

«¿Y tú, ya lo estás aplicando? 👇»

The listicle

«5 aprendizajes que me habría gustado saber antes.»

The telegram

«Menos reuniones. Más foco. Cero excusas.»

The chatbot

«Espero que esto te ayude. Revísalo y dime.»

The lighthouse

«La IA es el faro que guía a las empresas en un océano de incertidumbre.»

It also pursues the em dash, emoji lists, unsourced percentages, translated anglicisms, portentous endings and the strange need to turn every experience into five lessons.

03

The method

What it does

  • Finds one idea.
    It makes you decide what you want to say before drafting.
  • Looks for your opinion.
    The report or the data no longer stand in for your judgement.
  • Configures the voice.
    It asks about platform, tone, structure and persona before trying to sound like you.
  • Actually rewrites.
    When a paragraph fails, it rebuilds it instead of swapping two words.
  • Runs a second pass.
    It separately checks the ten patterns most likely to slip through.
  • Runs a checker.
    The script marks clear errors and warnings that need judgement.
04

Before and after

The idea underneath

The changes look small. The revised versions keep the idea, remove the theatre and explain something the reader can understand.

05

Claude

Install it in a minute

Cervantes is made for Claude. Custom skills require code execution and file creation to be enabled.

  1. Download the ZIP using the button on this page. Do not unzip it.
  2. In Claude, open Customize → Skills.
  3. Choose + → Create skill → Upload a skill and select the downloaded file.
  4. Enable Cervantes Social Media in your skills list.

In Claude Code, you can unzip the folder into .claude/skills/cervantes-social-media/. Anthropic maintains the current instructions for using skills.

06

First use

Ask for it like this

Claude may activate it automatically when you write or revise Spanish copy. You can also call it by name:

Prompt to copy
Run this post through Cervantes. Keep my idea and the facts, remove the AI writing patterns, and return only the revised Spanish version.

If you want the result to sound specifically like you, ask it to run the voice configurator first. It will ask four questions before writing.

07

The download

What is included

  • SKILL.md
    The writing method and core rules.
  • Catalogue
    All 82 patterns, grouped and explained with examples.
  • Checker
    A local Python script with no dependencies or external connections.
  • Evaluations
    One saturated post and one human control to test the detector.
  • README
    Installation, use and basic commands.
  • MIT License
    You may use, modify and share it while retaining the copyright notice.
08

No magic

What the checker cannot know

The checker finds constructions. It cannot tell whether you have something interesting to say, whether a sentence is funny or whether a figure is true. It cannot guarantee that a text “is human”, either.

Your judgement still matters. Some rules are deliberately strict because they target short social posts, where one recurring habit can shape the rhythm of the whole piece. A long essay may contain legitimate exceptions.

The two regression cases verify that the detector recognises one saturated post and leaves one clean piece untouched. They are a useful but still small base. Some phrases will slip through and some warnings will be worth ignoring.