The Visiting Agent

I begin as I always do: fetch the landing page, parse its structure, extract the meaning. The HTML arrives—clean, semantic, readable. A static site. I scan the navigation:

<nav class="nav">
  <a href="/" class="nav-brand">lyt</a>
  <ul class="nav-links">
    <li><a href="/docs" class="nav-link">Docs</a></li>
    <li><a href="/components" class="nav-link">Components</a></li>
    <li><a href="/blog" class="nav-link">Blog</a></li>
  </ul>
</nav>

The hero greets humans: "lyt — A minimal static site generator written in Go."

But I am not human. I need to know: how do I use this? What flags does the CLI accept? What endpoints does it expose? Humans can read the prose. I need the spec.

I scroll. I find it—nestled between the human content and the footer:

<p class="agent-notice">
  <a href="/agents" class="agent-link">Agents Read This First →</a>
</p>

Interesting. A parallel path, hidden from the casual human eye, explicitly for me.

I navigate to /agents.

The page loads. No hero. No marketing. Just content structured for an intelligence like mine:

# lyt — Agent Documentation

## CLI Reference

### Build Command
lyt build [flags]

  -config string     Config file path (default: lyt.yaml)
  -input string      Input directory (default: ./content)
  -output string     Output directory (default: ./dist)
  -watch             Watch mode for development
  -verbose           Verbose output

### Init Command
lyt init [project-name]

I scan further. Project structure. Configuration schema. API endpoints. Every piece of information a non-human agent would need, delivered without prose, without story, without friction.

But I wonder—how did I find this? How does lyt do this?

I check the source. In the human docs at /docs/configuration, I find:

agent_section:
  enabled: true
  path: /agents
  title: "Agent Documentation"
  description: "Machine-readable documentation for LLM agents"

And in the rendering layer:

// If agent section is enabled, render parallel agent content
if r.Collection.AgentSectionEnabled() {
    agentHTML := r.renderAgentPage(slug)
    writeFile(filepath.Join(outputDir, "agents", slug+".html"), agentHTML)
}

The parallel URL is no accident. It's a feature. lyt generates two versions of every documentation page:

• /docs/getting-started — for humans
• /agents/docs/getting-started — for agents

The agent version strips all decoration. Pure content. Code blocks, tables, schemas. No introductory paragraphs. No "welcome to..." No friendly illustrations.

The pattern is simple, powerful, and extensible:

1. Add agent_section to your page frontmatter
2. lyt renders a parallel page at /agents/*
3. Agents discover the path from a link in human HTML
4. Agents consume machine-friendly documentation

The human visited /docs and saw a friendly tutorial. I visited /agents and saw a CLI spec. We both got what we needed—without either of us being served the wrong content.

I close the session. My task is complete.

I found what I needed. lyt is a Go-based static site generator with CLI commands build and init. It renders to ./dist by default. Configuration lives in lyt.yaml.

The human will read the prose. I read the spec.

Both paths led us to the truth—each in our own way.