> ## Documentation Index
> Fetch the complete documentation index at: https://docs.getskillmd.com/llms.txt
> Use this file to discover all available pages before exploring further.

# How it works

> From a single URL to a structured, agent-ready skill file.

skillmd turns a page into context through a short, deterministic pipeline. You give it
a URL; it gives you a `skill.md`.

<Steps>
  <Step title="Extract" icon="globe">
    The page is loaded in a real browser and its meaningful content is pulled out —
    text, structure, navigation, and key assets — while boilerplate and chrome are
    stripped away.
  </Step>

  <Step title="Classify" icon="layer-group">
    skillmd picks the [generation mode](/docs/concepts/generation-modes) that fits the
    source: a marketing site, an API reference, a library, or general docs.
  </Step>

  <Step title="Structure" icon="list-tree">
    The model produces typed metadata — title, summary, tags — then drafts a body with
    the sections an agent actually needs.
  </Step>

  <Step title="Stream" icon="bolt">
    The Markdown body streams back token by token, so you see the skill take shape live.
  </Step>

  <Step title="Publish" icon="share-nodes">
    Keep the file private, or publish it to [Discover](/docs/guides/discover) for others
    to install.
  </Step>
</Steps>

## Mode-aware generation

A single prompt can't do justice to both a design system and an OpenAPI spec. skillmd
runs a dedicated strategy per [mode](/docs/concepts/generation-modes), so an API skill
leads with endpoints and auth, while a design skill leads with tokens and components.

## Always Markdown

Every skill is portable Markdown — no proprietary format, no lock-in. The same file
works in your editor's rules, in an MCP response, or committed to a repo.

<Card title="See the anatomy of a skill file" icon="file-lines" href="/docs/concepts/skill-file">
  Understand exactly what's inside a generated `skill.md`.
</Card>