# Docs for AI agents (/agents)



These docs are built for AI agents as well as people. Every page is available
as clean Markdown, and the whole site is indexed in machine-readable form. Each
section below stands on its own.

## Fetch any page as Markdown [#fetch-any-page-as-markdown]

Append `.md` to any docs URL to get its Markdown source — no HTML, no
navigation chrome:

```bash
curl https://docs.harmont.dev/pipeline-sdk/patterns.md
curl https://docs.harmont.dev/cli/run.md
```

## Or use content negotiation [#or-use-content-negotiation]

Send an `Accept: text/markdown` header and the same URL returns Markdown. Most
coding agents (e.g. Claude Code, Cursor) send this automatically:

```bash
curl https://docs.harmont.dev/cli/run -H 'Accept: text/markdown'
```

The response carries `Vary: Accept`, so a browser at the same URL still gets HTML.

## Site indexes [#site-indexes]

* [/llms.txt](https://docs.harmont.dev/llms.txt) — a Markdown index of every
  page, with links and one-line descriptions.
* [/llms-full.txt](https://docs.harmont.dev/llms-full.txt) — every page's
  Markdown concatenated into one document, for loading the whole corpus at once.

## Point an agent at Harmont [#point-an-agent-at-harmont]

The fastest way to get a correct pipeline from an agent: tell it to read
[/pipeline-sdk/patterns.md](https://docs.harmont.dev/pipeline-sdk/patterns.md)
first. That page is the canonical do/don't guidance — chiefly, **use the
high-level toolchain APIs (`hm.rust.toolchain()`, `hm.cmake()`, …) rather than
hand-rolling commands through the general-purpose `hm.sh(...)` escape hatch.**

For a working starting point, copy an [example](/examples) and adapt it.