# Llms Txt > Escribe y valida un archivo llms.txt — la convención /llms.txt que da a los asistentes de IA un mapa curado del sitio en vez de dejarles adivinar a partir del HTML.4 Fuente: https://skillsagentes.com/skills/kraayenjon/claude-pagekit/llms-txt Markdown: https://skillsagentes.com/skills/kraayenjon/claude-pagekit/llms-txt.md Repositorio: https://github.com/kraayenjon/claude-pagekit Autor: kraayenjon Licencia: NOASSERTION Actualizado: hace 14 horas Coste de contexto: 120 tok instalada, 1.8k tok al activarse, 1.8k tok con todos los archivos del bundle Bundle: 1 archivo, 7 KB Permisos que pide: ninguno declarado ## Instalación Un skill son archivos markdown: los mismos archivos valen para cualquier agente y lo único que cambia es el directorio de destino, es decir la bandera `--agent`. Añade `-g` para instalarlo en todos los proyectos de la máquina. ```bash # Claude Code npx -y skills add kraayenjon/claude-pagekit --skill llms-txt --agent claude-code # Cursor npx -y skills add kraayenjon/claude-pagekit --skill llms-txt --agent cursor # Codex npx -y skills add kraayenjon/claude-pagekit --skill llms-txt --agent codex # Gemini CLI npx -y skills add kraayenjon/claude-pagekit --skill llms-txt --agent gemini # Windsurf npx -y skills add kraayenjon/claude-pagekit --skill llms-txt --agent windsurf # Cline npx -y skills add kraayenjon/claude-pagekit --skill llms-txt --agent cline ``` ## Qué hace - Escribe un fichero /llms.txt con el formato exacto (H1, blockquote, secciones ## y enlaces con descripción) - Decide qué páginas incluir y cuáles excluir (archivos, paginación, login, checkout, etc.) - Genera el compañero opcional /llms-full.txt con el contenido completo en Markdown limpio - Indica dónde colocar los ficheros según el framework y cómo enlazarlos desde robots.txt y el - Valida un llms.txt existente contra una checklist de formato y veracidad ## Cuándo usarla - El usuario menciona 'llms.txt', 'llms-full.txt' o 'AI crawler file' - El usuario quiere que la IA entienda mejor su sitio o que ChatGPT lo lea bien - Se está preparando un sitio para ser citado por asistentes de IA ## Qué la activa - "Ayúdame a crear el archivo llms.txt de mi web" - "Necesito un llms-full.txt para mi documentación" - "Revisa si mi llms.txt está bien formateado" - "Quiero que ChatGPT entienda mejor mi sitio, ¿cómo hago un AI crawler file?" ## Antes de instalar - makes network requests ## Archivos - SKILL.md — 7 KB ## SKILL.md Reproducido tal cual desde kraayenjon/claude-pagekit bajo NOASSERTION. Esta sección es el documento original y está en inglés. # llms.txt A single Markdown file at `/llms.txt` that tells an AI assistant what a site is and where the good pages are. It exists because HTML is noisy — navigation, cookie banners, footers, scripts — and a model reading a page has to guess what matters. `llms.txt` removes the guessing. It is a convention, not a standard. No crawler is required to read it. Write one anyway: it costs ten minutes, it is read by a growing number of assistants and tools, and the act of writing it forces you to state plainly what the site is for. ## The format Strict Markdown, in this order. The structure is the contract — a parser looks for exactly this shape. ```markdown # Site or business name > One sentence saying what this is. Concrete. No marketing adjectives. Optional free prose. A short paragraph or two of context that does not fit a link — what the business does, who it serves, where it operates, what makes it different. Keep it factual. ## Section name - [Page title](https://example.com/page): what a reader gets from this page - [Another page](https://example.com/other): one clause, not a sentence ## Optional - [Lower-priority page](https://example.com/archive): things a model can skip if short on space ``` Rules that actually matter: 1. **One `# H1`**, the name. First line of the file. 2. **A `>` blockquote directly after it**, one sentence. This is the line most likely to be quoted back verbatim, so write it like a caption, not a slogan. 3. **`##` sections** group the links. Name them for what they are — `Services`, `Case studies`, `Docs`, `About` — not `Resources` or `More`. 4. **Every list item is a link followed by `:` and a description.** The description is the point. `[Pricing](/pricing): three plans, from €19/month, no setup fee` is useful. `[Pricing](/pricing): our pricing page` is noise. 5. **Absolute URLs.** A model may read the file with no knowledge of the origin. 6. **A section literally named `## Optional`** is the one piece of magic: anything under it is explicitly marked as skippable when context is tight. Use it. Do not use it for anything you actually want read. ## What goes in Only pages a person would be glad to land on. A good `llms.txt` for a small site is ten links. For a large one it is thirty, not three hundred. Include: the home page, what you sell or do, pricing, about, contact, the three or four best pieces of writing, documentation entry points, anything with real specifics in it. Leave out: tag and category archives, pagination, login and account pages, checkout, thank-you pages, anything `noindex`, anything thin, and anything duplicated across URLs. ## The one-page case A single landing page still gets an `llms.txt`, and it is mostly prose rather than links. The blockquote and the paragraph under it carry everything: ```markdown # Panadería Aurora > A wood-fired sourdough bakery in Malasaña, Madrid, open Tuesday to Sunday from 8am. Aurora bakes four breads daily — country sourdough, seeded rye, olive fougasse and a Saturday-only brioche — in a wood oven installed in 1974. Everything is sold from the counter at Calle del Pez 14; there is no delivery and no online ordering. Wholesale enquiries from restaurants are taken by phone. ## Pages - [Home](https://panaderiaaurora.es/): breads, hours, address, and how to order wholesale ## Optional - [Instagram](https://instagram.com/panaderiaaurora): daily bake announcements ``` That is a complete, correct file. Do not pad it. ## llms-full.txt The optional companion at `/llms-full.txt` holds the actual **content** — the full text of the important pages, concatenated as clean Markdown, so an assistant can read the substance without fetching anything. Write one when the site is documentation, a knowledge base, or anything where the value is in the text. Skip it for a one-page landing site: `llms.txt` already contains everything. Keep it plain Markdown: `# Page title`, then the page's prose. No navigation, no HTML, no scripts, no repeated header and footer. Strip everything a reader would skip. ## Where the files go Both live at the domain root, served as `text/plain` or `text/markdown`: | Framework | Location | |---|---| | Next.js (App Router) | `public/llms.txt` | | Astro, Vite, SvelteKit | `public/llms.txt` | | Hugo, Jekyll, Eleventy | `static/llms.txt` | | Plain static site | next to `index.html` | Verify after deploying: `https://yourdomain.com/llms.txt` must return the raw file, not an HTML 404 page. Then point `robots.txt` at it, so a crawler that does not know the convention still finds it: ``` User-agent: * Allow: / User-agent: GPTBot Allow: / User-agent: ClaudeBot Allow: / User-agent: PerplexityBot Allow: / User-agent: Google-Extended Allow: / Sitemap: https://yourdomain.com/sitemap.xml ``` `llms.txt` is not a `robots.txt` directive, so it has no line of its own there. Linking it from the home page `` is harmless and occasionally helps: ```html ``` ## Validating one Read it back and check: - [ ] Exactly one `#` heading, and it is the first line - [ ] A `>` blockquote immediately after it, one sentence, factual - [ ] Every `##` section has at least one item - [ ] Every list item is `- [text](absolute-url): description` - [ ] No relative URLs - [ ] No link to a page that 404s, redirects, or is `noindex` - [ ] Every description says something specific — no "our services page" - [ ] Under about 100 links; if longer, most of them belong under `## Optional` or nowhere - [ ] Nothing in it is untrue. A model will repeat this verbatim The last one is the important one. `llms.txt` is the most quotable file on a site — everything in it is written in a model's preferred format, pre-chewed and unambiguous. An invented claim here does not get buried the way it would in a paragraph of HTML. It gets repeated. ## Writing the blockquote It is one sentence and it does most of the work. Three failure modes: | Bad | Why | Better | |---|---|---| | "Innovative solutions for modern businesses." | Says nothing. Fits any company. | "Payroll software for Spanish companies with 5–50 employees." | | "The best bakery in Madrid." | Unverifiable claim, reads as marketing, gets discounted. | "A wood-fired sourdough bakery in Malasaña, Madrid." | | "We are a full-service creative agency offering branding, web design, motion graphics, packaging and strategy for clients across sectors." | Too long, no shape, nothing to lift. | "A three-person branding studio in Lisbon working mostly with restaurants and hotels." | Name the thing, the who, and the where. Numbers and place names survive summarisation; adjectives do not. ## Dónde encaja - Categoría: [SEO y GEO](https://skillsagentes.com/categorias/seo-geo.md) — Keywords, auditorías on-page, datos estructurados y visibilidad en respuestas de IA. - Creador: [kraayenjon](https://skillsagentes.com/creators/kraayenjon.md) — 0 skills en el directorio - [Todas las skills](https://skillsagentes.com/skills.md) - [Ranking de instalaciones](https://skillsagentes.com/ranking.md) ## Otras skills del mismo repositorio - [Landing Es](https://skillsagentes.com/skills/kraayenjon/claude-pagekit/landing-es.md): Escribe copy de landing en español que suene escrito en español, no traducido: tú/usted/vos, CTAs que convierten, vocabulario de España vs Latinoamérica, formatos de precio/fecha/teléfono y páginas legales obligatorias.este es - [Pagekit](https://skillsagentes.com/skills/kraayenjon/claude-pagekit/pagekit.md): Crea una landing page o portafolio completo a partir de una entrevista breve, pensado para alguien que no programa: entrevista, diseño, scaffolding con Next.js + Tailwind + shadcn/ui, despliegue sin cuenta de Vercel y SEO/AEO listos. ## Skills relacionadas - [Landing Es](https://skillsagentes.com/skills/kraayenjon/claude-pagekit/landing-es.md): Escribe copy de landing en español que suene escrito en español, no traducido: tú/usted/vos, CTAs que convierten, vocabulario de España vs Latinoamérica, formatos de precio/fecha/teléfono y páginas legales obligatorias.este es - [Pagekit](https://skillsagentes.com/skills/kraayenjon/claude-pagekit/pagekit.md): Crea una landing page o portafolio completo a partir de una entrevista breve, pensado para alguien que no programa: entrevista, diseño, scaffolding con Next.js + Tailwind + shadcn/ui, despliegue sin cuenta de Vercel y SEO/AEO listos. - [Seo Unlighthouse](https://skillsagentes.com/skills/agricidaniel/claude-seo/seo-unlighthouse.md): Auditoría Lighthouse multi-página vía el CLI de código abierto Unlighthouse. Alternativa gratuita a correr PageSpeed contra cada URL del sitio, sin consumir cuota de API, corre en local. - [Seo Image Gen](https://skillsagentes.com/skills/agricidaniel/claude-seo/seo-image-gen.md): Generación de imágenes con IA para assets SEO: OG/social preview, imágenes hero de blog, imágenes de schema, fotografía de producto e infografías. Usa Gemini vía el pipeline banana Creative Director. - [Seo Flow](https://skillsagentes.com/skills/agricidaniel/claude-seo/seo-flow.md): Integración del framework FLOW: SEO basado en evidencia con el ciclo Find → Leverage → Optimize → Win. Expone 41 prompts de IA por etapa desde la base de conocimiento FLOW (CC BY 4.0). --- Skills Agentes · [Índice de páginas en markdown](https://skillsagentes.com/sitemap.md) · [Inicio](https://skillsagentes.com/index.md)