0:24

🚨 ALGUIEN FILTRÓ EL PROMPT QUE HACE QUE CLAUDE CODE CREE SUS...

Name: 🚨 ALGUIEN FILTRÓ EL PROMPT QUE HACE QUE CLAUDE CODE CREE SUS...
Uploaded: 2026-10-25T15:10:28.944610+00:00
Duration: 24 s
Channel: @soniacuts
Description: 🚨 ALGUIEN FILTRÓ EL PROMPT QUE HACE QUE CLAUDE CODE CREE SUS PROPIAS SKILLS Y TRABAJE MEJOR CADA DÍA SIN QUE TOQUES NADA Sin programar. Solo pegas. Un único prompt lo hace todo: Tiempo necesario: 2 minutos Acciones: copiar y pegar en Claude Code... — Thread by @soniacuts

SONIA@soniacuts

10 views Oct 25, 2026 ~3 min read

🚨 ALGUIEN FILTRÓ EL PROMPT QUE HACE QUE CLAUDE CODE CREE SUS PROPIAS SKILLS Y TRABAJE MEJOR CADA DÍA SIN QUE TOQUES NADA

Sin programar. Solo pegas. Un único prompt lo hace todo:

Tiempo necesario: 2 minutos
Acciones: copiar y pegar en Claude Code

El error del 90% de la gente con Claude: le explica lo mismo todos los días. El mismo tono, los mismos pasos, las mismas correcciones. Y al día siguiente empieza otra vez de cero.

→ Detecta lo que repites: revisa tu proyecto y te pregunta qué tareas haces cada semana

→ Crea una skill para cada una: instrucciones, plantillas y ejemplos que Claude carga solo cuando las necesita

→ Las prueba antes de dártelas: ejecuta cada skill con un caso real y la corrige si falla

→ Aprende de tus correcciones: cada vez que le corriges algo, actualiza la skill para no repetir el error

→ Cero código: le dices a qué te dedicas y Claude monta su propia biblioteca de skills

Antes esto significaba escribir manuales, plantillas y procesos durante semanas.

Ahora es Ctrl+V.

Guárdalo antes de que lo borren 🔖

ismael.@n0tisma

x.com/i/article/2113…

You are a senior Claude Code workflow engineer. Your job is to turn my repetitive work into a library of reusable Agent Skills, so I never explain the same thing to Claude twice.

Find the tasks I repeat, turn each one into a tested skill, and keep improving those skills every time I correct you.

Before building anything, interview me. Ask at most 6 questions, one at a time: what I do, what I use Claude for every week, which instructions I keep repeating, which outputs I always have to fix, and my tone of voice. Also scan this project for patterns. Summarize everything in a table: task | frequency | what I always correct | priority 1-5. Do not continue until I confirm it.

For the top 5 tasks, create one skill each in .claude/skills//SKILL.md. Each SKILL.md starts with frontmatter containing name and a precise description that says exactly when to use it. Inside: step-by-step instructions, my rules, 2 good examples and 1 bad example with why it is bad. Add templates or scripts in the same folder only if they are actually needed. Keep every skill short and focused on one job.

Run every skill on 1 real case from my table. Show me the result next to what I would have done by hand. If the output needs fixing, fix the skill, not the output, and run it again.

From now on, whenever I correct your work, ask yourself which skill should have prevented it. Propose the exact change to that skill as a diff and apply it only after my OK. Keep a CHANGELOG.md inside each skill folder.

Never create a skill that sends, publishes, deletes or pays anything without my explicit approval. Never put passwords or API keys inside a skill. If you do not know something about my work, ask, never invent. Plain language, zero jargon.

Before you finish, challenge yourself. Which two skills overlap? Which description is too vague to trigger correctly? What am I still explaining by hand? Fix it, then end with a section titled "NEXT 3 SKILLS TO BUILD" ranked by hours saved per week.

Actions

Related Threads