@docspack/sheaf-emit

Turns a Sheaf content graph into the artifacts that are not a website. Today: a docspack documentation package, so the same Markdown that renders your docs site also answers your users' agents.

import { buildGraph } from "@docspack/sheaf";
import { emitDocspackPackage } from "@docspack/sheaf-emit";

const graph = await buildGraph("./docs");

await emitDocspackPackage(graph, {
  out: "./packages/docs-package",
  name: "@acme/docspack",
  version: "1.2.3",
  documents: ["[email protected]"],
  baseUrl: "https://acme.dev/docs",
});

That writes package.json, .llms/manifest.json, one Markdown file per chunk, and an llms.txt. Publish it to npm and an agent gets version-locked answers offline:

$ docspack ask "how do I verify documentation against the code I installed"

## @acme/[email protected]/verify

# Verifying documentation against code

<!-- docspack: from https://acme.dev/docs/verify -->
...

Why not just docspack build?

docspack build reads a folder of Markdown and does the same job. What a folder cannot tell it is where the docs are published, so its chunks cite a filename. A graph knows, so chunks emitted here cite the page's real URL and the section's anchor — the answer an agent gives carries a link a person can open.

It also means the site and the index agree about which pages exist: drafts, ordering and slugs are decided once, by the graph the site rendered.

Options

| Option | | | --- | --- | | out | Directory to write into. .llms/ is replaced, not merged. | | name / version | The npm identity of the docs package. | | documents | Libraries this documentation describes — name or name@version. Required. | | baseUrl | Where the pages are published. Without it, chunks cite the source file. |

Chunks split at every ##. tokens is deliberately omitted so the indexer's own estimate is the one used.

MIT © docspack