Agent Skills
Agent Skills are instructions for AI assistants: each skill is a SKILL.md with frontmatter describing what an assistant can do with your product and the steps to do it, optionally with any number of reference files.
Movk Nuxt Docs uses nuxt-agent-discovery to scan the skills/ directory at the project root at build time, generate an index and publish the files under /.well-known/skills/, following the Agent Skills Discovery convention.
Publish a skill
Create the directory
Create skills/ at the project root (next to nuxt.config.ts), with one subdirectory per skill:
my-docs/
├── nuxt.config.ts
└── skills/
└── my-skill/
└── SKILL.md
Write SKILL.md
description is required. State what the skill does and when to use it; assistants decide when to invoke it from this text:
---
name: my-skill
description: |
Write component docs for My Library.
Use when asked to "document a component" or "describe the props".
---
# My Skill
## Workflow
1. Read the component source and list its props, slots and emits
2. Generate the page from references/template.md
Build and deploy
The build log lists the skills it found (Found 1 agent skill: my-skill); skills failing validation are skipped with a warning. After deploying they are served at:
curl https://docs.example.com/.well-known/skills/index.json
curl https://docs.example.com/.well-known/skills/my-skill/SKILL.md
Install
Readers install every skill the site publishes with the skills CLI, which detects the assistants installed locally and writes to each:
npx skills add https://docs.example.com
Naming rules
The skill name is the frontmatter name, or the directory name when omitted. It must:
- Be 1–64 characters of lowercase letters, digits and hyphens
- Not start or end with a hyphen, nor contain consecutive hyphens (
--) - Match its directory name exactly
Layout
A skill directory may hold any supporting files, which assistants load on demand. Files and directories starting with . and symbolic links are not published.
skills/
└── my-skill/
├── SKILL.md
├── references/
│ └── template.md
└── assets/
└── checklist.md
Every file is listed in index.json and served at /.well-known/skills/{name}/{path}. Keep SKILL.md under about 500 lines and move details into references/ to save the assistant's context.
A site can publish several skills; the index is sorted by name and assistants pick one from its description.
Change the directory
The scanned directory is agentDiscovery.skills.dir, relative to the project root and skills by default in the layer. Set skills to false to stop publishing skills:
export default defineNuxtConfig({
agentDiscovery: {
skills: {
dir: 'agent-skills'
}
}
})
Built-in skills
This site publishes two skills that work for any project using Movk Nuxt Docs:
| Skill | Purpose |
|---|---|
create-docs | Analyze a project and generate a complete Movk Nuxt Docs site: layout, configuration, pages and AI integration |
review-docs | Review docs for structure, clarity, SEO and technical accuracy, with Critical / Important / Suggested findings |
npx skills add https://docs.mhaibaraai.cn
Compared with llms.txt
| llms.txt | Agent Skills | |
|---|---|---|
| Content | A directory of pages and links | Actionable workflows, constraints and examples |
| Loaded | When the assistant looks something up | On demand, when the skill is triggered |
| Purpose | Help the assistant find information | Teach the assistant to use your product |
They complement each other: llms.txt tells an assistant where the information is, a skill tells it what to do.
MCP Server
Every docs site serves a Model Context Protocol server at /mcp by default. This page lists the built-in tools, resources and prompts, shows how to connect Claude Code, Cursor, VS Code and other clients, and explains how to disable, restrict or extend the server.
llms.txt
The docs site generates /llms.txt and /llms-full.txt automatically, giving language models an index and the full text of your docs. Learn where the defaults come from, how to set titles and sections, how they relate to each page's Markdown, and how to use them in Cursor and other tools.